CN117934159A

CN117934159A - Personal credit report query monitoring and early warning method based on artificial intelligence

Info

Publication number: CN117934159A
Application number: CN202410324640.1A
Authority: CN
Inventors: 黎乾术; 黄瑞; 张翔
Original assignee: Beijing Xinli Hechuang Information Technology Co ltd
Current assignee: Beijing Xinli Hechuang Information Technology Co ltd
Priority date: 2024-03-21
Filing date: 2024-03-21
Publication date: 2024-04-26

Abstract

The invention relates to the technical field of credit monitoring, in particular to an artificial intelligence-based personal credit report query monitoring and early warning method. In the invention, the dynamic credit scoring model is combined with the current economic index and market data to predict, so that credit state analysis is more suitable for market change, timeliness and correlation are improved, a group intelligent algorithm enables the whole credit behavior of the market to be accurately simulated, macroscopic market trend grasp is enhanced, a simulated annealing algorithm provides flexibility in personal differential financial scene analysis, a clustering analysis and pattern recognition method carefully analyzes consumption behavior and credit preference, a comprehensive customer image is constructed, and a feature extraction method analyzes user behavior features and social network influence through t-distribution random neighborhood embedding, so that a new view angle is provided for accurate marketing and risk management.

Description

Personal credit report query monitoring and early warning method based on artificial intelligence

Technical Field

The invention relates to the technical field of credit monitoring, in particular to a personal credit report query monitoring and early warning method based on artificial intelligence.

Background

Credit monitoring involves a key area of financial security and personal privacy protection. Credit monitoring techniques are primarily focused on tracking and analyzing personal credit reports in order to timely mine any anomalies or potential fraud. In this field, technological developments aim to help individuals and financial institutions reduce credit risk and prevent identity theft by effectively monitoring and analyzing the individuals credit history.

The personal credit report inquiry monitoring and early warning method can automatically check personal credit reports, monitor credit activities and send early warning when any abnormal or suspicious transaction is found. The main purpose of this approach is to protect consumers from credit fraud and identity theft, while helping them maintain good credit status. By periodically monitoring the credit report, the individual can know any change of the own credit record in time, and the accuracy and the legality of all information are ensured.

The traditional personal credit report query monitoring method has obvious defects. On credit score trend analysis, the recognition capability of a long-term periodic pattern is lacking, and the comprehensiveness of risk prediction is limited. Failure to fully utilize real-time economic indicators and market data results in a lack of dynamic adaptability in credit status predictions. In the analysis of market credit behaviors, the support of a group intelligent algorithm is lacking, and macroscopic market changes are difficult to comprehensively reflect. Personal financial context analysis does not refer to personal differentiation factors, reducing predictive personalization and accuracy. The customer behavior analysis mainly adopts a simple statistical method, and fails to construct a comprehensive customer portrait, so that customer relationship management and market policy effectiveness are affected. These deficiencies lead to risk management blindness and market opportunity misses in practice, limiting credit management and market competition effectiveness.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides an artificial intelligence-based personal credit report query monitoring and early warning method.

In order to achieve the above purpose, the present invention adopts the following technical scheme: an artificial intelligence-based personal credit report query monitoring and early warning method comprises the following steps:

s1: based on personal historical credit data, adopting a time sequence analysis algorithm to analyze credit score trend and periodicity pattern, predicting potential risks and generating credit trend analysis results;

s2: based on the credit trend analysis result, a dynamic credit scoring model is adopted, and the prediction analysis of the future credit state is carried out by combining the current economic index and market data, so that the future credit prediction result is generated;

s3: based on the future credit prediction result, adopting a group intelligent algorithm to perform simulation analysis on the overall credit behavior of the market, and referring to the prediction of key credit events to generate a market credit behavior simulation result;

S4: based on the market credit behavior simulation result, analyzing the personal differentiated financial scene by adopting a simulated annealing algorithm, predicting credit scores under various situations, and generating a personal financial scene analysis result;

S5: based on personal financial transaction data and the personal financial scene analysis result, adopting a cluster analysis and pattern recognition method to analyze the consumption behavior and credit preference of the client and constructing a comprehensive client portrait;

S6: based on the comprehensive customer portrait, adopting a feature extraction method, analyzing the behavior features of the user and the influence of the social network through t-distribution random neighborhood embedding, and generating a behavior feature analysis report;

S7: based on the behavior characteristic analysis report, a network analysis method is adopted to construct a network model of personal credit transaction, a network structure is analyzed, the network structure comprises node centrality and cluster coefficients, key nodes and potential risk connection are identified, and a credit network structure analysis result is generated;

s8: based on the credit network structure analysis result, the network data is analyzed by adopting a machine learning technology, an abnormal transaction mode and potential credit fraud are identified, and a credit risk identification report is generated.

As a further aspect of the present invention, the credit trend analysis results include a trend graph, a periodic pattern graph, and a risk index list, the future credit prediction results include a prediction credit score graph and a risk inter-area estimation, the market credit behavior simulation results include a market trend prediction graph and a key event list, the personal financial scenario analysis results include a scenario analysis graph and a credit score prediction, the comprehensive customer representation includes a consumption behavior pattern graph, a credit preference classification, and a risk classification, the behavior feature analysis report includes a behavior feature graph, a consumption habit analysis, and a social network impact assessment, the credit network structure analysis results include a network topology graph, a key node identification, and a risk connection analysis, and the credit risk identification report includes a risk transaction pattern identification, a fraud prediction, and a risk early warning index.

As a further scheme of the invention, based on personal historical credit data, a time sequence analysis algorithm is adopted to analyze credit score trend and periodicity pattern, and predict potential risks, and the step of generating credit trend analysis results is specifically as follows:

S101: based on personal historical credit data, adopting an autoregressive moving average model, and referring to the existing credit score value to predict the change trend of the future score, and simultaneously evaluating the volatility and long-term trend in the data to generate basic trend prediction;

S102: based on the basic trend prediction, decomposing the time series data into trend components, seasonal components and residual components by adopting a seasonal decomposition time series analysis method, revealing periodic changes of credit scores, and generating seasonal change refinements;

s103: based on the seasonal variation refinement, adopting an isolated forest algorithm, isolating data points by constructing a plurality of decision trees, identifying abnormal points and potential risks in the data, and generating abnormal point identification;

s104: and integrating the basic trend prediction, seasonal variation refinement and abnormal point identification results, adopting an integrated analysis method, and providing multidimensional credit score trend assessment by comparing and fusing multiple analysis results to generate a credit trend analysis result.

As a further scheme of the invention, based on the credit trend analysis result, a dynamic credit scoring model is adopted, and the current economic index and market data are combined to carry out the prediction analysis of the future credit state, and the step of generating the future credit prediction result comprises the following steps:

S201: based on the credit trend analysis result, adopting a linear regression model to analyze historical credit data, including calculating a linear relation between historical credit scores and future trends, and predicting future credit score changes by analyzing the relation between variables in a time sequence to generate historical data correlation analysis;

S202: based on the historical data correlation analysis, carrying out macro economic index analysis by adopting a vector autoregressive model, wherein the analysis comprises the steps of analyzing the influence of economic indexes of interest rate and failure rate on credit scores, and generating macro economic index analysis by predicting a combined effect through processing a plurality of time series data;

S203: based on the macro economic index analysis, a neural network model is adopted to conduct market trend influence analysis, the influence of market data comprising stock market indexes and consumer confidence on personal credit scores is analyzed, the relation between market changes and the credit scores is captured through the network model, and the market trend influence analysis is generated;

S204: based on the market trend influence analysis, comprehensive prediction analysis of the future credit state is carried out by adopting a multiple regression model, the influence of historical data, macro economy and market trend is integrated, and a future credit prediction result is provided through multivariate analysis.

As a further scheme of the invention, based on the future credit prediction result, adopting a group intelligent algorithm to carry out simulation analysis on the overall credit behavior of the market, and referring to the prediction of key credit events, the steps of generating the market credit behavior simulation result are specifically as follows:

s301: based on the future credit prediction result, adopting a particle swarm optimization algorithm, and adjusting the individual position to capture the optimal solution by simulating collective behaviors to reflect the dynamic change of the credit behaviors under the differentiated market conditions so as to generate individual credit behavior dynamic simulation;

s302: based on the dynamic simulation of the credit behaviors of the individuals, adopting a genetic algorithm to simulate the credit behaviors of the whole market, optimizing the credit behavior mode of the market based on natural selection and genetic mechanism, predicting the credit variation trend of the whole market, and generating a credit behavior evolution simulation of the market;

S303: based on the market credit behavior evolution simulation, a system dynamics method is adopted to analyze the influence of critical events, particularly financial crisis or policy change, on the market credit behavior, and the propagation and influence of macroscopic events in the market are simulated by constructing a dynamic model of the market credit behavior to generate a critical event market influence simulation;

S304: and integrating the individual credit behavior dynamic simulation, the market credit behavior evolution simulation and the key event market influence simulation results, adopting a group intelligent algorithm to perform the simulation analysis of the whole market credit behavior, and generating a market credit behavior simulation result by referring to the influence of the key credit event.

As a further scheme of the invention, based on the market credit behavior simulation result, a simulated annealing algorithm is adopted to analyze the personal differentiated financial scene and predict credit scores under various situations, and the step of generating the personal financial scene analysis result specifically comprises the following steps:

S401: based on the market credit behavior simulation result, performing global optimization simulation of personal credit score by adopting a simulated annealing algorithm, capturing an optimal solution through random search to optimize the credit score of the individual under the condition of differentiated market, and generating a credit score global optimization simulation;

S402: based on the credit score global optimization simulation, adopting a decision tree algorithm to analyze credit score decisions of individuals under the target financial scene, and distinguishing credit score results under the differentiated financial scene by constructing a classification tree to generate credit score decision analysis;

S403: based on the credit score decision analysis, carrying out conditional probability analysis of personal credit scores by adopting a Bayesian network, and providing probability prediction for the credit scores in each scene by calculating the credit score probability in the differential financial scene to generate credit score probability prediction;

S404: and comprehensively simulating the credit score global optimization, analyzing the credit score decision and predicting the credit score probability, adopting a simulated annealing algorithm to comprehensively analyze the credit scores of individuals under various financial situations, predicting the credit scores under different situations, and generating an analysis result of the personal financial situations.

As a further scheme of the invention, based on personal financial transaction data and the personal financial scene analysis result, a cluster analysis and pattern recognition method is adopted to analyze the consumption behavior and credit preference of the customer, and the steps of constructing the comprehensive customer portrait are specifically as follows:

S501: based on personal financial transaction data and the personal financial scene analysis result, grouping consumption behavior data by adopting a K-means clustering algorithm, and dividing consumption behaviors into a plurality of categories according to a distance minimization principle by calculating the distance between each data point and a clustering center to generate a consumption behavior clustering result;

S502: based on the consumption behavior clustering result, key features are extracted from consumption behavior data by adopting principal component analysis, the data is converted into a group of linear independent variables through orthogonal transformation, the data dimension is reduced, key information is highlighted, and key consumption feature extraction is generated;

S503: based on the key consumption feature extraction, classifying and pattern recognition are carried out on the extracted features by adopting a support vector machine, differentiated consumption patterns are distinguished by capturing an optimal hyperplane, and consumption pattern classification analysis is generated;

s504: and integrating the consumption behavior clustering result, the key consumption characteristic extraction and the consumption pattern classification analysis result, and carrying out multidimensional analysis on the consumption behavior and the credit preference of the client by adopting a pattern recognition method to construct an integrated client portrait.

As a further scheme of the invention, based on the comprehensive customer portrait, a feature extraction method is adopted, and the behavior features and social network influence of the user are analyzed through t-distribution random neighborhood embedding, and the steps of generating a behavior feature analysis report are specifically as follows:

S601: based on the comprehensive customer portrait, carrying out mathematical transformation on the features in the original data set by adopting a principal component analysis algorithm, projecting the data into a new coordinate system, identifying key variables in the data, and extracting feature vectors and feature values of the covariance matrix by calculating the covariance matrix of the data to generate a key user feature set;

S602: based on the key user feature set, performing dimension reduction processing on the feature set by adopting a t-distribution random neighborhood embedding algorithm, and capturing optimal representation of similarity in a low-dimensional space through calculating similarity probability between features in a high-dimensional space, so as to reduce data dimension and generate a dimension reduction user feature map;

S603: based on the dimension reduction user feature map, analyzing the position of a user in a social network by adopting network centrality analysis, and generating a user social network influence map by calculating the centrality and the betweenness centrality of each node;

S604: based on the user social network influence diagram and the dimension reduction user characteristic diagram, integrating chart information by adopting a data fusion technology, performing pattern recognition and association analysis on the integrated data by adopting a statistical analysis method, analyzing the relationship between the user behavior characteristics and the social network positions of the user, and generating a behavior characteristic analysis report.

As a further scheme of the invention, based on the behavior feature analysis report, a network analysis method is adopted to construct a network model of personal credit transaction, analyze a network structure, including node centrality and cluster coefficients, identify key nodes and potential risk connection, and generate a credit network structure analysis result specifically as follows:

S701: based on the behavior feature analysis report, constructing a personal credit transaction network model by adopting a social network analysis algorithm, and using a scaleless network model and a small world network model, reflecting network characteristics by calculating the connection probability and reconnection probability between nodes to generate a credit transaction network model;

S702: based on the credit transaction network model, carrying out node centrality analysis by adopting a centrality analysis method in graph theory, identifying influence nodes in the network by calculating the direct connection number of each node, evaluating the importance of the betweenness centrality by quantifying the bridging action of the nodes in the network, and generating a node centrality analysis report;

S703: based on the credit transaction network model, a cluster coefficient calculation method is applied to analyze the cluster trend of the network, a local cluster coefficient method is used for identifying a tightly connected node group by calculating the connection density between adjacent nodes of a target node, and a global cluster coefficient method is used for evaluating the overall connectivity of the network to generate a cluster coefficient analysis report;

S704: and combining the node centrality analysis report and the cluster coefficient analysis report, determining strategic nodes in the network by using a key node and risk connection identification method according to comprehensive centrality and betweenness centrality results, revealing risk links by analyzing abnormal transaction modes among the nodes, and generating a credit network structure analysis result.

As a further scheme of the present invention, based on the analysis result of the credit network structure, the machine learning technology is adopted to analyze the network data, identify the abnormal transaction mode and the potential credit fraud, and generate the credit risk identification report specifically includes:

S801: based on the credit network structure analysis result, adopting a principal component analysis algorithm to reduce the dimensionality of the network data, reducing redundant information by extracting key feature vectors of the data, and reserving key variability of the data to generate optimized network data;

S802: based on the optimized network data, adopting a decision tree algorithm to perform feature selection on the data, screening classification features by creating a tree structure of a decision rule, and performing pattern recognition on the screened features by using a random forest algorithm to generate feature selection and pattern recognition results;

S803: based on the feature selection and pattern recognition result, an isolated forest algorithm is applied to analyze an abnormal transaction pattern, abnormal points are isolated by randomly selecting features and randomly segmenting feature values, the abnormal pattern is detected, a support vector machine is used to analyze, a decision boundary surrounding normal data points in a feature space is constructed, the abnormal points are separated, and an abnormal transaction pattern analysis result is generated;

S804: and integrating and explaining the analysis result by adopting a data visualization and interpretation model according to the analysis result of the abnormal transaction mode, improving the intuitiveness of the analysis result by using a graphic representation method, and generating a credit risk identification report by providing a logic and basis of a model decision by using the interpretation model.

Compared with the prior art, the invention has the advantages and positive effects that:

According to the invention, the accuracy of credit score trend and periodic pattern analysis is improved by applying a time sequence analysis algorithm, so that potential risks are effectively predicted. The dynamic credit scoring model is combined with the current economic index and market data to predict, so that credit state analysis is more suitable for market change, and timeliness and relevance are improved. The group intelligent algorithm enables the whole credit behavior of the market to be accurately simulated, and the grasping of macroscopic market trend is enhanced. The simulated annealing algorithm provides flexibility in personal differential financial scenario analysis, enhancing personalized services. The cluster analysis and pattern recognition method carefully analyzes consumption behaviors and credit preferences, builds a comprehensive customer portrait, and enhances the management efficiency of the customer relationship. The feature extraction method analyzes the user behavior features and social network influence through t-distribution random neighborhood embedding, and provides a new view for accurate marketing and risk management.

Drawings

FIG. 1 is a schematic diagram of the main steps of the present invention;

FIG. 2 is a detailed schematic of the S1 of the present invention;

FIG. 3 is a schematic diagram of an S2 refinement of the present invention;

FIG. 4 is a schematic diagram of an S3 refinement of the present invention;

FIG. 5 is a schematic diagram of an S4 refinement of the present invention;

FIG. 6 is a schematic diagram of an S5 refinement of the present invention;

FIG. 7 is a schematic diagram of an S6 refinement of the present invention;

FIG. 8 is a schematic diagram of an S7 refinement of the present invention;

Fig. 9 is a schematic diagram of the S8 refinement of the present invention.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Example 1:

referring to fig. 1, the present invention provides a technical solution: an artificial intelligence-based personal credit report query monitoring and early warning method comprises the following steps:

S2: based on credit trend analysis results, a dynamic credit scoring model is adopted, and prediction analysis of future credit states is carried out by combining current economic indexes and market data, and future credit prediction results are generated;

s3: based on a future credit prediction result, adopting a group intelligent algorithm to carry out simulation analysis on the overall credit behavior of the market, and referring to the prediction of key credit events, generating a market credit behavior simulation result;

s4: based on the market credit behavior simulation result, analyzing the personal differential financial scene by adopting a simulated annealing algorithm, predicting credit scores under various situations, and generating a personal financial scene analysis result;

S5: based on personal financial transaction data and personal financial scene analysis results, adopting a cluster analysis and pattern recognition method to analyze the consumption behavior and credit preference of the client and constructing a comprehensive client portrait;

S7: based on the behavior feature analysis report, a network analysis method is adopted to construct a network model of personal credit transaction, a network structure is analyzed, the network structure comprises node centrality and cluster coefficients, key nodes and potential risk connection are identified, and a credit network structure analysis result is generated;

s8: based on the analysis result of the credit network structure, the network data is analyzed by adopting a machine learning technology, abnormal transaction modes and potential credit fraud behaviors are identified, and a credit risk identification report is generated.

The credit trend analysis results comprise a change trend graph, a periodic pattern graph and a risk index list, the future credit prediction results comprise a prediction credit score graph and a risk interval estimation, the market credit behavior simulation results comprise a market trend prediction graph and a key event list, the personal financial scene analysis results comprise a scene analysis graph and a credit score prediction, the comprehensive customer portraits comprise a consumption behavior pattern graph, credit preference classification and risk level division, the behavior feature analysis report comprises a behavior feature graph, consumption habit analysis and social network influence assessment, the credit network structure analysis results comprise a network topology graph, key node identification and risk connection analysis, and the credit risk identification report comprises risk transaction pattern identification, fraud behavior prediction and risk early warning indexes.

In a first step, algorithms such as ARIMA (autoregressive integral moving average) are mainly used, based on a time series analysis of the historical credit data of the individual, which models predict future trends from the historical credit data. The algorithm first removes the non-stationarity of the data by a differential method, and then captures the dependency and random fluctuations of the time series using an Autoregressive (AR) and Moving Average (MA) model. In this process, the algorithm determines the optimal parameters of the model, such as the hysteresis order, to minimize the prediction error. Furthermore, the analysis of the periodic patterns involves detecting seasonal components in the data, implemented using seasonal differences.

In the second step, a dynamic credit scoring model based on credit trend analysis results is combined with economic indicators and market data to predict the future credit state, and algorithms such as regression analysis, random forest or neural network are generally adopted. The algorithms can construct a comprehensive model to predict the future credit status of the individual according to the credit trend analysis result of the individual and combining macro economic indicators (such as GDP growth rate, loss rate and the like) and market data (such as stock market index, interest rate and the like). The model may be trained and validated to optimize its predictive performance. The output result is a future credit forecast report, which illustrates possible changes in personal credit scores under differentiated economic conditions, providing data support for credit decisions.

And thirdly, adopting a group intelligent algorithm to carry out simulation analysis on the whole credit behaviors of the market. In step, common algorithms include Particle Swarm Optimization (PSO) or Genetic Algorithms (GA) that can simulate the behavior and interactions of a large number of individuals in the credit market. The model simulates the response and overall market trend of market participants by setting different market scenarios and parameters, such as credit policy changes, economic fluctuations, etc. In addition, the prediction of critical credit events relies on analysis of historical data and market dynamics to predict critical events that may occur in the future by identifying past patterns and trends. The output of the steps is a market credit simulation report detailing predictions of market credit and key events in the case of differentiation.

And in the fourth step, analyzing the personal differential financial scene through a simulated annealing algorithm. Simulated annealing is a heuristic optimization algorithm that searches for optimal solutions by simulating the process of slow cooling after metal heating. In step, the algorithm finds the most likely credit score in different financial scenarios through a number of random sampling and iterative processes. The algorithm may consider various financial factors such as revenue level, liability status, and historical repayment behaviors. By this means, a change in credit score of an individual under different circumstances (e.g., economic decline, change in interest rate, etc.) can be predicted. The final yield is a personal financial situational analysis report providing credit score predictions in various contexts, providing a reference for personal financial planning and credit management.

In a fifth step, customer consumption behavior and credit preferences are analyzed by cluster analysis and pattern recognition methods based on personal financial transaction data and financial context analysis results. Cluster analysis identifies patterns of consumption by grouping similar transaction behaviors, while pattern recognition techniques such as Support Vector Machines (SVMs) or neural networks are used to identify specific patterns of consumption and credit behaviors. These methods may reveal individual consumption habits, preferences, and credit usage patterns. Through this analysis, an integrated customer representation can be constructed describing the financial behavior and credit preferences of the customer. The final output is a comprehensive customer portrait report, which provides deep customer behavior insight and provides basis for personalized financial services and credit product design.

And sixthly, analyzing the user behavior characteristics and social network influence by a characteristic extraction and t-distribution random neighborhood embedding (t-SNE) method. Feature extraction involves identifying and selecting features that are most representative of and predictive of customer behavior, such as transaction frequency, shopping type, social interactions, etc. the t-SNE is an efficient dimension reduction technique that can efficiently represent high-dimensional data in a lower-dimensional space, revealing potential structures and patterns in the data. By this means, the behavioral characteristics of the user and their location in the social network and their impact can be analyzed. The final yield is a behavioral characteristic analysis report describing the behavioral characteristics of the user and roles in the social network, providing basis for targeted marketing strategies and credit risk assessment.

In the seventh step, a network analysis method is adopted to construct a network model of the personal credit transaction. In step, a network analysis tool, such as a graph theory algorithm, is used to analyze the structure of the personal credit transaction network. This includes calculating node centrality in the network to identify the most influential individuals in the credit network, and calculating cluster coefficients to assess how tight the population is in the network. In addition, it relates to identifying critical nodes and potentially risky connections in a network, such as frequent abnormal transaction patterns or potential credit fraud. The final output is a credit network structure analysis result, which shows the structure characteristics of the credit transaction network, key nodes and potential risks, and provides important information for credit risk management and fraud prevention.

And finally, analyzing network data by adopting a machine learning technology based on the credit network structure analysis result, and identifying abnormal transaction modes and credit fraud. Machine learning algorithms include decision trees, logistic regression, or deep learning models. These algorithms identify abnormal transaction patterns or fraud by analyzing network data such as transaction frequency, amount, relationship between the transaction parties, etc. The model is trained and tested on a large amount of historical transaction data to improve its identification accuracy. The final output is a credit risk identification report that illustrates the identified abnormal transaction patterns and potential credit fraud, providing an effective risk management tool for the financial institution.

Referring to fig. 2, based on personal historical credit data, a time sequence analysis algorithm is adopted to analyze credit score trend and periodicity pattern, and predict potential risks, and the steps of generating credit trend analysis result are specifically as follows:

S102: based on basic trend prediction, decomposing time series data into trend components, seasonal components and residual components by adopting a seasonal decomposition time series analysis method, revealing periodic variation of credit score, and generating seasonal variation refinement;

S103: based on seasonal variation refinement, adopting an isolated forest algorithm, isolating data points by constructing a plurality of decision trees, identifying abnormal points and potential risks in the data, and generating abnormal point identification;

In a substep S101, personal historical credit data is analyzed by an autoregressive moving average (ARMA) model. The data format is typically time-series data, i.e. chronological personal credit scores. First, the algorithm analyzes autocorrelation in the time series, i.e., the relationship between the current value and its past value, by means of an auto-regression (AR) section. For example, the score of the current month is predicted using the credit score of the past three months. The Moving Average (MA) portion is then used to model random error terms of the time series, smoothing random fluctuations in the historical data. In the process, the model will select the optimal number of lag terms and moving average terms to minimize the prediction error. Thereafter, the model predicts a trend of change in the future credit score based on the historical data and evaluates the volatility and long-term trends in the data. Finally, the step generates a basic trend prediction report, which shows the predicted trend and volatility evaluation of the future credit score in detail, and provides an important basis for credit risk management.

In the S102 substep, the periodic variation of the credit score is refined by a time series analysis method of seasonal decomposition. In the step, time-series data is first decomposed into three parts: trend component, seasonal component, and residual component. The trend component shows a long-term trend of the credit score over time, and the seasonal component reveals the influence of a periodic factor such as holidays or seasonal variations on the credit score. Finally, the residual component contains random fluctuations in addition to trending and seasonal factors. The decomposition process is typically implemented using a moving average or other filtering technique. By this decomposition, the different modes of credit variation over time, especially those periodic fluctuations, can be seen more clearly. The output of the steps is a seasonal variation refinement report that provides an analysis of the periodic variation of credit score that aids in better understanding and predicting credit behavior.

In S103 substep, outliers and potential risks in the data are identified using an orphan forest algorithm. The isolated forest algorithm isolates data points by constructing a plurality of decision trees, each of which randomly selects a feature and a score value to segment the data. This method is particularly suitable for detecting outliers, which tend to be easily isolated and can be separated from other data with fewer divisions. In processing the credit score data, the algorithm may effectively identify those score values that differ significantly from most data, which may be indicative of credit risk or fraud. By the method, an abnormal point identification report can be generated, and abnormal points and possible risk factors in credit data are listed in detail, so that the method is important for credit monitoring and risk prevention.

Finally, in the sub-step S104, the results of the previous steps are combined, and a multi-dimensional credit score trend assessment is provided by using an integrated analysis method. This step involves comprehensive comparison and fusion of the results of the basic trend prediction, seasonal variation refinement, and outlier identification. By this integrated analysis, a more comprehensive and thorough credit score trend assessment can be obtained. For example, the integrated analysis may reveal correlations between long-term trends in credit scores, seasonal fluctuations, and potential points of abnormal risk. Finally, the credit trend analysis report generated by the steps provides a multi-angle view for credit assessment, which helps financial institutions to more accurately assess and manage credit risk.

Consider a set of simulated personal credit time series data, such as the credit score of the last 12 months: [680, 700, 710, 720, 730, 740, 750, 760, 770, 780, 790, 800]. In step S101, these data are analyzed using an ARMA model, which may identify a trend of steadily rising credit score month by month. In step S102, the seasonal decomposition reveals a significant increase or decrease in credit score in certain months (e.g., holiday seasons). In step S103, the orphan forest algorithm identifies that the credit score for a month is abnormally low (e.g., a month suddenly drops to 650), which is a potential risk signal. Finally, in step S104, the analysis results are combined to generate a comprehensive credit trend analysis report, which provides a comprehensive view of personal credit score variation trends, periodic fluctuations, and potential risk points.

Referring to fig. 3, based on the credit trend analysis result, the steps of performing prediction analysis of the future credit state by adopting a dynamic credit scoring model and combining the current economic index and market data, and generating the future credit prediction result are specifically as follows:

S201: based on the credit trend analysis result, adopting a linear regression model to perform historical credit data analysis, including calculating a linear relation between historical credit scores and future trends, and predicting future credit score changes by analyzing the relation between variables in a time sequence to generate historical data correlation analysis;

S202: based on historical data correlation analysis, carrying out macroscopic economic index analysis by adopting a vector autoregressive model, wherein the analysis comprises the steps of analyzing the influence of economic indexes of interest rate and failure rate on credit scores, and generating macroscopic economic index analysis by predicting a combined effect through processing a plurality of time series data;

S203: based on macro economic index analysis, a neural network model is adopted to conduct market trend influence analysis, the influence of market data including stock market indexes and consumer confidence on personal credit scores is analyzed, the relation between market changes and the credit scores is captured through the network model, and market trend influence analysis is generated;

s204: based on market trend influence analysis, comprehensive prediction analysis of the future credit state is carried out by adopting a multiple regression model, the influence of historical data, macro economy and market trend is integrated, and a future credit prediction result is provided through multivariate analysis.

In the sub-step S201, the historical credit data is analyzed by a linear regression model, and quantization of the linear relationship between the historical credit score and the future trend is achieved. The specific operation process comprises data preprocessing, model construction, parameter optimization and result interpretation. In the data preprocessing stage, the primary task is to collect and scrub historical credit data, which includes extracting credit and associated time stamps from a credit database. The data cleaning work involves missing value processing, outlier rejection and the like, and ensures the accuracy and the integrity of data. Next, in the model building stage, a linear regression equation y=ax+b is used, where Y represents the future credit score, X represents the time variable, and a and b are coefficients of the linear regression model. And calculating the most suitable a and b values through least square method and other technologies, so that the model can accurately predict the future credit score. In the parameter optimization stage, the model parameters are adjusted by using the technologies of cross verification, grid search and the like so as to improve the accuracy of prediction. Finally, in the result interpretation stage, regression coefficients and statistical test results output by the model are analyzed to evaluate the strength and direction of the relationship between the historical credit score and the future trend. After completion, the generated report not only reveals the correlation of the historical data, but also provides a quantitative basis for future trends in the credit score.

In the sub-step S202, macro economic index analysis is performed by a vector autoregressive model, and in-depth analysis of interactions between economic indexes such as interest rate, failure rate and the like and credit scores is performed. The operational steps of this stage involve data preparation, model setup, causal relationship verification, and result interpretation. In the data preparation stage, macro economic data such as interest rate and loss rate are collected, and time alignment is carried out on the macro economic data and credit scoring data, so that consistency of data time sequences is ensured. The model setting stage uses a Vector Autoregressive (VAR) model that can process multiple time series data simultaneously and capture the dynamic relationship between them. The VAR model reveals the impact of economic indicators on credit scores by estimating the interdependencies between different time sequences. Next, in the causality checking stage, granger causality checking methods are used to verify whether the interest rate and the loss rate are the Granger causes of credit scores. Finally, in the result interpretation stage, the output results of the VAR model, such as the impulse response function and variance decomposition, are analyzed to understand how the macro economic indicators affect the credit score. The final result of the analysis is a report revealing the correlation between the macro economic indicators and the credit scores and providing a predictive basis for future trends in the credit scores.

In the step S203, a market trend impact analysis is performed by a neural network model, so as to explore the impact of market data such as stock market indexes, consumer confidence, etc. on personal credit scores. The operation flow comprises data set construction, neural network design, model training and result analysis. The data set construction stage requires collecting market trend related data, such as stock market index, consumer confidence index, and combining these data with credit scoring data to form a data set for training the neural network. In the neural network design stage, a proper neural network architecture, such as a multi-layer perceptron or a convolutional neural network, is selected according to the data characteristics, and the layer number, the neuron number, the activation function and the like of the network are determined. In the model training stage, network parameters are adjusted through methods such as a back propagation algorithm, gradient descent and the like so as to minimize prediction errors. And in the result analysis stage, the performance of the neural network is evaluated by using the test data set, and how the complex relationship between the market change and the credit score is captured by the network model is analyzed. The generated analysis report provides insight into the impact of market trends on credit scores, helping to better resolve interactions between market dynamics and personal credits.

In the S204 substep, comprehensive prediction analysis of the future credit status is performed by a multiple regression model, aiming at integrating the influence of historical data, macro economy and market trend and providing future credit prediction. This step includes data integration, model construction, model evaluation, and predictive application. In the data integration stage, historical credit data, macro economic indexes and market trend data are fused together to form a comprehensive data set. The model construction stage uses a multiple regression model that allows multiple predicted variables to enter the model simultaneously, capturing the relationship between these variables and the credit score. And in the model evaluation stage, the fitting degree and the prediction capability of the model are evaluated through statistical indexes such as R square, R square adjustment, AIC and the like. And in the prediction application stage, predicting the credit score in the future by using the constructed multiple regression model, and generating a prediction report. This report not only demonstrates the predicted outcome of future credit status, but also provides a comprehensive analysis of the possible influencing factors of the personal credit score, with significant reference value for credit assessment and management.

Referring to fig. 4, based on the future credit prediction result, a group intelligent algorithm is adopted to perform simulation analysis on the overall credit behavior of the market, and the step of generating the market credit behavior simulation result by referring to the prediction of the key credit event is specifically as follows:

s301: based on a future credit prediction result, adopting a particle swarm optimization algorithm, and adjusting individual positions to capture an optimal solution by simulating collective behaviors to reflect dynamic changes of credit behaviors under a differentiated market condition so as to generate individual credit behavior dynamic simulation;

S302: based on the dynamic simulation of the credit behaviors of individuals, adopting a genetic algorithm to simulate the credit behaviors of the whole market, optimizing the credit behavior mode of the market based on natural selection and genetic mechanism, predicting the credit variation trend of the whole market, and generating a market credit behavior evolution simulation;

S303: based on market credit behavior evolution simulation, a system dynamics method is adopted to analyze the influence of critical events, particularly financial crisis or policy variation, on the market credit behavior, and the propagation and influence of macroscopic events in the market are simulated by constructing a dynamics model of the market credit behavior to generate a critical event market influence simulation;

S304: and integrating the individual credit behavior dynamic simulation, the market credit behavior evolution simulation and the key event market influence simulation results, adopting a group intelligent algorithm to perform the simulation analysis of the whole market credit behavior, and referring to the influence of the key credit event to generate a market credit behavior simulation result.

In the step S301, the process of dynamic simulation of the credit behaviors of the individuals is completed through a particle swarm optimization algorithm. First, in the data preparation phase, it is necessary to collect individual credit history data including credit scores, debit records, repayment histories, and the like. The data format is typically in a structured tabular form, facilitating algorithmic processing. Next, a population of particles is initialized, where each particle represents one possible credit behavior solution, whose position and velocity are randomly initialized. Particle swarm optimization algorithm simulates the prey behavior of the bird swarm, and each particle adjusts the flight path according to own experience and swarm experience. Specifically, each particle in the algorithm will update its own position and velocity based on two "optimal" positions: one is the optimal location (individual optimum) where the particle itself has been captured so far, and the other is the optimal location (global optimum) where the entire population of particles has been captured so far. These optimal locations are determined by an evaluation function (e.g., a credit risk assessment model) that quantifies the effect of the credit action scheme represented by each particle. As the iterative process proceeds, particle swarms tend to capture optimal solutions in solution space, i.e., schemes that best represent future credit behavior. Finally, the generated dynamic simulation reflects possible changes in the individual credit behaviors under differentiated market conditions, providing a valuable perspective for resolving and predicting individual credit behaviors.

In the S302 substep, the credit behaviors of the whole market are simulated through a genetic algorithm, and the market credit behavior evolution simulation is carried out. The steps are based on market-level credit data, such as credit score distribution of differentiated groups, market lending behavior trends, and the like. The data format is also a structured table. Genetic algorithms are search algorithms that simulate the natural evolution process, optimizing problem solutions through selection, crossover and mutation operations. In this process, an initial population is first generated, each individual representing a possible solution to market credit behavior. The fitness of each individual is evaluated by a fitness function (e.g., an evaluation index based on market stability and credit health). In the selection stage, individuals with high fitness are selected for crossover and mutation, and new individuals are generated. Crossover refers to two individuals exchanging part of their genes, while mutation refers to random alteration of individual genes, thereby introducing new properties. As the iteration proceeds, individuals with higher fitness will dominate the population, thereby gradually developing an optimal market credit behavior pattern. The finally generated market credit behavior evolution simulation reveals the development trend of the market credit behavior, and has important significance for understanding the credit dynamics of the whole market.

In the sub-step S303, the influence of macroscopic events such as financial crisis or policy change on market credit behavior is simulated by a system dynamics method. The data used in the steps comprise macro economic indexes, policy change records and the like, and are in various formats, including time series data and text descriptions. System dynamics is a method for simulating complex system behaviors by constructing a dynamic model reflecting the system structure and feedback mechanisms. In this step, a kinetic model is first constructed that includes various economic indicators and credit behavior variables. The model includes a plurality of differential equations describing interactions between the differencing variables. Next, the effects of macroscopic events (such as financial crisis occurrences or policy changes) on the variables in the model are simulated, observing how these events propagate in the market and affect credit behavior. Through iterative computation of the time step, the dynamic model shows the change trend of market credit behavior along with time. The finally generated simulation results reveal the influence mechanism and development trend of the key macroscopic events on the market credit behaviors.

In the S304 substep, the simulation results of the previous step are integrated, and the integrated simulation analysis of the market credit behaviors is performed through a group intelligent algorithm. This step requires the integration of individual credit data, market credit evolution data, and critical event impact data, the format of which includes structured data and simulation result data. The swarm intelligence algorithm is used to integrate information at different levels in this step to discover market credit patterns. The algorithm finds the optimal solution by modeling interactions between individuals in the population. Specifically, the algorithm analyzes patterns and trends in the differential data sources, and gradually builds a model capable of comprehensively reflecting market credit behaviors through information exchange among individuals in an iterative process. This model not only incorporates individual and market level information, but also refers to the impact of macroscopic events. The final generated market credit behavioral simulation results provide a comprehensive and deep view of resolving the dynamics of the entire credit market.

Referring to fig. 5, based on the market credit behavior simulation result, a simulated annealing algorithm is adopted to analyze the personal differentiated financial scenario, and predict credit scores under various situations, and the steps of generating the personal financial scenario analysis result are specifically as follows:

S401: based on market credit behavior simulation results, performing global optimization simulation of personal credit scores by adopting a simulated annealing algorithm, capturing an optimal solution through random search to optimize the credit scores of individuals under the condition of differentiated markets, and generating a global optimization simulation of the credit scores;

S402: based on credit score global optimization simulation, adopting a decision tree algorithm to analyze credit score decisions of individuals under target financial scenes, and distinguishing credit score results under differentiated financial scenes by constructing a classification tree to generate credit score decision analysis;

S403: based on credit score decision analysis, a Bayesian network is adopted to perform conditional probability analysis of personal credit scores, and probability prediction is provided for the credit scores in each scene by calculating the credit score probability in the differential financial scene, so as to generate credit score probability prediction;

S404: comprehensive credit score global optimization simulation, credit score decision analysis and credit score probability prediction are performed, a simulated annealing algorithm is adopted to perform credit score comprehensive analysis of individuals under various financial situations, credit scores under different situations are predicted, and a personal financial situation analysis result is generated.

In the sub-step S401, a global optimization simulation of personal credit score is performed by a simulated annealing algorithm, and an iterative method is used to capture the optimal solution in the complex market credit behavior simulation. First, the input data is personal credit history and related market behavior data, which typically includes a person's loan record, repayment history, credit card usage, and the like. At the beginning of the simulated annealing algorithm, a high "temperature" is set at which the algorithm randomly selects points in the solution space as starting points and calculates its score. The scoring function evaluates credit scores based on the market model and the personal behavior data. As the iterative process proceeds, the temperature gradually decreases, the algorithm tries to generate a new solution at each step, and calculates a score for the new solution. Accepting the new solution if the score of the solution is higher; if the score is low, there is also a probability of acceptance to avoid local optima. This process is repeated until the temperature drops to a preset threshold, at which point a globally optimal or near optimal solution is deemed found. In this way, the personal credit score can be optimized in different market environments to more accurately reflect the personal credit status. The process generates a globally optimized personal credit score model that enables more accurate predictions and evaluations of personal credit scores for different market conditions.

In a sub-step S402, the person' S credit score decisions in the target financial scenario are analyzed by a decision tree algorithm. This process involves taking as input the person's financial data and credit records to build a classification tree. Data includes, but is not limited to, revenue levels, liability conditions, historical credit behavior, and the like. In constructing the decision tree, the root node is first determined, which represents the most important decision factor. The algorithm then splits based on the features in the dataset to form branch and leaf nodes. Each branch represents a decision path, while leaf nodes represent the final credit scoring result. In each step of splitting, the algorithm selects the feature that most reduces data uncertainty, and uses criteria such as information gain or genie uncertainty to select the splitting attribute. Finally, the process forms a complete classification tree that can divide and predict personal credit scores based on different financial scenarios. By the method, the variation trend of the personal credit score under different financial conditions can be clearly seen, and a reliable analysis basis is provided for credit score decision. This process generates a detailed credit score decision analysis report that shows the possible outcomes of personal credit scores and their probabilities in different financial scenarios.

In a substep S403, a conditional probability analysis of the personal credit score is performed through the bayesian network. In this step, the input data includes financial information of the individual, historical credit data, and market scenario data. A bayesian network is a graphical model that represents probabilistic relationships between variables through nodes and directed edges. Each node represents a variable, such as revenue, expense, history of expiration, etc., and the edges represent dependencies between the variables. By setting up the network structure and conditional probability tables, the algorithm is able to calculate the conditional probability of certain variables (e.g., credit scores) given the other variables (e.g., specific financial scenarios). This process involves using statistical methods to estimate parameters in conditional probability tables and applying bayesian reasoning for probability computation. Finally, the analysis method can provide probability prediction for credit score under each financial scenario, and provides a probability-based view for credit score decision. This process generates a report containing predictions of credit score probabilities for various financial scenarios that aid in understanding the possible variations in credit scores under different conditions.

In the sub-step S404, the previous global optimization simulation, credit score decision analysis and credit score probability prediction are combined, and credit score comprehensive analysis of the individual under various financial situations is performed through a simulated annealing algorithm. The process again utilizes a simulated annealing algorithm, but the input data at this time is a comprehensive data set including the optimized credit score, decision tree analysis results, and probabilistic predictions of the bayesian network. By means of an iterative search optimal solution method, various different financial situations are comprehensively considered by the algorithm, and a scoring model capable of most accurately reflecting personal credit conditions is found. In this process, the setting of temperature parameters and the selection of annealing strategies are critical, which determine the efficiency of the search process and the accuracy of the results. This process ultimately creates a personal credit score model that integrates various financial scenarios, which can provide a more comprehensive and thorough credit score analysis. The comprehensive analysis result is a comprehensive report containing personal credit score prediction under different financial situations, and has important significance for credit management and personal financial planning.

Referring to fig. 6, based on personal financial transaction data and personal financial scenario analysis results, a cluster analysis and pattern recognition method is adopted to analyze consumption behavior and credit preference of a customer, and the steps of constructing a comprehensive customer portrait are specifically as follows:

S501: based on personal financial transaction data and personal financial scene analysis results, grouping consumption behavior data by adopting a K-means clustering algorithm, dividing consumption behaviors into a plurality of categories according to a distance minimization principle by calculating the distance between each data point and a clustering center, and generating consumption behavior clustering results;

S503: based on key consumption feature extraction, classifying and pattern recognition are carried out on the extracted features by adopting a support vector machine, differentiated consumption patterns are distinguished by capturing an optimal hyperplane, and consumption pattern classification analysis is generated;

S504: and integrating the consumption behavior clustering result, the key consumption characteristic extraction and the consumption pattern classification analysis result, and performing multidimensional analysis on the consumption behavior and credit preference of the client by adopting a pattern recognition method to construct an integrated client portrait.

In the S501 substep, the individual' S financial transaction data is grouped for consumption behavior by a K-means clustering algorithm. The process first involves preprocessing of the data, including normalization of information of the amount consumed, time of consumption, class of consumption, etc., to ensure that the data is comparable in different dimensions. The K-means clustering algorithm is based on this data, and K data points are randomly selected as initial clustering centers at the beginning. Next, the algorithm iteratively performs the steps of: first, the distance from each data point to each cluster center is calculated, and the data points are distributed to the nearest cluster center to form clusters. The center point of each cluster, i.e., the mean of all points within the cluster, is then recalculated. This process is repeated until the cluster center is no longer significantly changed or a preset number of iterations is reached. The key to the K-means algorithm is the calculation of the distance and the updating of the cluster center, typically using euclidean distance as a metric. In this way, the consumption behavior data of an individual is effectively divided into a plurality of categories, each category representing a particular consumption pattern. The consumption behavior clustering result generated in the process reveals the diversity and regularity of the individual consumption behaviors, and provides a basis for subsequent analysis.

In the sub-step S502, key feature extraction is performed on the clustering result of the consumption behavior by Principal Component Analysis (PCA). The data input for this process is clustered consumption behavior data including, but not limited to, multi-dimensional information of consumption amount, time, frequency, etc. Key to principal component analysis is the conversion of raw data into a set of linearly independent variables, principal components, by orthogonal transformation. The algorithm first calculates the covariance matrix of the data, and then finds out the eigenvalues and eigenvectors of the covariance matrix. The eigenvectors define a new coordinate system, and the magnitude of the eigenvalues determines the importance of each principal component. The data is re-represented in terms of these principal components, with the first few principal components typically selected as key features to reduce the data dimension and preserve the most variability. The key consumption features extracted by PCA capture the most important patterns and trends in consumption behavior data, and provide a data basis for reducing noise and strengthening key information for deep consumption pattern analysis.

In the S503 substep, the extracted key consumption features are classified and pattern-identified by a Support Vector Machine (SVM). SVM is a powerful supervised learning method for classification and regression analysis of data. In this step, the input data is the key consumer feature derived from the PCA. The goal of the SVM is to find an optimal hyperplane that maximizes the boundaries between different classes. The algorithm determines the position and orientation of the hyperplane by solving an optimization problem that aims to maximize the boundary interval. SVM also utilizes kernel skills to process non-linearly separable data by mapping the data into a high-dimensional space to become linearly separable. In this step, the SVM classifies the individual's consumption behavior according to the key consumption features, thereby distinguishing between different consumption patterns. The consumption pattern classification analysis result generated in the process reveals the deep pattern and tendency of the personal consumption behavior, and provides important analysis basis for customer portraits and personalized services.

In the sub-step S504, the results of the consumer behavior clustering result, the key consumer feature extraction and the consumer pattern classification analysis are integrated, and the consumer behavior and the credit preference of the consumer are subjected to multidimensional analysis by a pattern recognition method. This step involves the integrated analysis of the data generated in the first three steps using a variety of data analysis techniques including statistical analysis, machine learning algorithms, etc. The goal is to build a comprehensive customer representation that includes not only the consumer behavior characteristics of the customer, but also reflects the credit preferences and potential risks of the customer. In the process, the algorithm comprehensively refers to various factors such as consumption habits, purchasing power, credit history and the like of the clients, and recognizes the correlation between the behavior characteristics and the credit characteristics of the clients through pattern recognition. The comprehensive customer representation generated by this process is a multidimensional data model that provides a means for financial institutions to gain insight into customers, facilitating credit assessment, marketing strategy formulation, and risk management.

Referring to fig. 7, based on the comprehensive customer portrait, the steps of analyzing the behavioral characteristics and social network influence of the user by t-distributed random neighborhood embedding and generating a behavioral characteristic analysis report are specifically as follows:

s602: based on a key user feature set, performing dimension reduction processing on the feature set by adopting a t-distribution random neighborhood embedding algorithm, and capturing optimal representation of similarity in a low-dimensional space by calculating similarity probability between features in a high-dimensional space, so as to reduce data dimension and generate a dimension reduction user feature map;

s603: based on the dimension-reduction user feature map, analyzing the position of a user in a social network by adopting network centrality analysis, and generating a user social network influence map by calculating the degree centrality and the betweenness centrality of each node;

S604: based on the user social network influence diagram and the dimension reduction user feature diagram, integrating chart information by adopting a data fusion technology, performing pattern recognition and association analysis on the integrated data by adopting a statistical analysis method, analyzing the relationship between the user behavior feature and the social network position, and generating a behavior feature analysis report.

In the sub-step S601, the original data features are mathematically transformed using a Principal Component Analysis (PCA) algorithm by integrating the data sets of the customer representation. First, each feature in the original dataset is normalized to ensure that the mean of each feature is 0 and the variance is 1. Normalization is an important preprocessing step because PCA is very sensitive to the scale of the data. The covariance matrix of the data is then calculated, which is a matrix that exhibits a degree of linear correlation between the features. The extraction of eigenvalues and eigenvectors of the covariance matrix is the core of the PCA. The eigenvalues represent the magnitude of the variance of the data in the direction of the corresponding eigenvector, which defines the new coordinate system. The feature vectors are arranged according to the descending order of the feature values, and the first several feature vectors are selected to form a new feature space. In this way, the data is projected into the new coordinate system, thereby achieving a reduction in dimensions. This process identifies key variables in the data, i.e., principal components, which are linear combinations of variables in the original dataset. The effect of this step is to generate a set of key user features that reduces the dimensionality of the data while retaining the most important information.

In the S602 substep, a t-distributed random neighborhood embedding (t-SNE) algorithm is adopted to perform dimension reduction processing based on the key user feature set. the t-SNE algorithm first calculates the probability of similarity between features in a high-dimensional space. This is typically achieved by computing gaussian joint probabilities for each pair of data points in a high dimensional space. Next, in the low-dimensional space, the algorithm tries to configure each point in such a way that their probability of similarity is as close as possible to that in the high-dimensional space. In this process, the algorithm uses a gradient descent method to optimize the locations of these low-dimensional points, thereby capturing structures in the high-dimensional data. In this way, the t-SNE is able to efficiently represent the local structure of high-dimensional data in a low-dimensional space, while also preserving some global structure. Finally, this step generates a reduced-dimension user profile that demonstrates the distribution of the data in a low-dimensional space, helping to identify patterns and populations in the data.

In the sub-step S603, based on the dimension-reduced user profile, a network centrality analysis is employed to analyze the user' S position in the social network. This step involves calculating the degree centrality and the betting centrality of each node. Centrality simply measures how many connections a node has, while mid-centrality measures how much a node mediates in the network, i.e., how many shortest paths through the node. These calculations reveal which users are centrally located in the social network and which users are bridging in connecting different social groups. Through the analysis, the generated user social network influence diagram not only shows the social network status of the user, but also reveals the structural characteristics in the social network.

In the S604 substep, integrating the chart information by adopting a data fusion technology in combination with the user social network influence diagram and the dimension-reduction user characteristic diagram. In the process, a statistical analysis method is used for carrying out pattern recognition and relevance analysis on the integrated data. These analyses aim at resolving the relationship between the user behavioral characteristics and their social network locations. For example, correlation analysis or regression models are used to explore the correlations between features. In this way, key behavioral characteristics that affect the user's social network status, or vice versa, how the social network status affects the user's behavior, can be identified. Finally, the behavioral profile report generated by this step provides a deep hole in the complex interactions between the user's behavior and its position in the social network.

Assume a set of user data sets containing age, income, online time duration, etc. In step S601, we may find age and income to be the main contributors through PCA. In S602, using t-SNE dimension reduction may reveal clusters of different ages and revenue groups. In S603, the network analysis may show that young users or high-income users are more active in the social network. Finally, in S604, the data fusion and analysis may indicate that young and high-income users are more influential in social networks and tend to exhibit specific patterns of behavior. These steps together constitute a comprehensive analysis flow that provides valuable insight by revealing deep patterns and trends in the user data.

Referring to fig. 8, based on a behavioral characteristic analysis report, a network analysis method is adopted to construct a network model of personal credit transaction, analyze a network structure, including node centrality and cluster coefficients, identify key nodes and potential risk connections, and generate a credit network structure analysis result specifically as follows:

S701: based on the behavioral characteristic analysis report, constructing a personal credit transaction network model by adopting a social network analysis algorithm, and using a scaleless network model and a small world network model, reflecting network characteristics by calculating connection probability and reconnection probability between nodes to generate a credit transaction network model;

S702: based on a credit transaction network model, carrying out node centrality analysis by adopting a centrality analysis method in graph theory, identifying influence nodes in a network by calculating the direct connection number of each node, evaluating the importance of the betweenness centrality by quantifying the bridging action of the nodes in the network, and generating a node centrality analysis report;

S703: based on a credit transaction network model, a cluster trend of the network is analyzed by applying a cluster coefficient calculation method, a node group which is tightly connected is identified by a local cluster coefficient method through calculating the connection density between adjacent nodes of a target node, the overall connectivity of the network is evaluated by a global cluster coefficient method, and a cluster coefficient analysis report is generated;

In the sub-step S701, a personal credit transaction network model is constructed by a social network analysis algorithm, and the specific implementation procedure using the scaleless network and the small world network model is as follows: first, personal credit transaction data, which may include personal basic information, transaction records, credit scores, etc., needs to be collected and consolidated. Next, a scaleless network model is used to build the network, the core of which is that most nodes have only a small number of connections, while few nodes have a large number of connections. Here, a real world credit transaction network is simulated with a scaleless network, where nodes represent individuals and edges represent credit transactions. In constructing a network, a probabilistic model will be applied to determine whether a connection is established between nodes, which typically depends on the degree of the node (i.e., the number of connections). Subsequently, a small world network model is employed to further refine the network structure. Small world networks have the characteristics of high cluster coefficients and short average path lengths, which means that in the network, a node is not only tightly connected to its neighbors, but can also reach any other node in the network with a small number of hops. In the model, these characteristics of the network are reflected by adjusting the reconnection probability (i.e., the probability that two nodes of an existing edge are disconnected and a new connection is established with other nodes in the network). The steps are fine-tuned by an algorithm, so that the generated credit transaction network model can reflect the actual credit transaction relationship among individuals and reveal the macroscopic structural characteristics of the network.

In the sub-step S702, the process of node centrality analysis by the centrality analysis method in graph theory involves two centralities: center of degree and center of median. The centrality analysis is performed by calculating the number of direct connections for each node in the network. Specifically, for each node, the number of other nodes directly connected to it is counted, and the higher this number is, the greater the influence of that node in the network is. To achieve this, the algorithm traverses each node in the network and calculates its degree. The betting center analysis is more complex and evaluates the importance of nodes by quantifying their bridging effect in the network. Specifically, the algorithm will calculate the frequency of occurrence of each node in all shortest paths. Here, the shortest path refers to a path connecting the minimum number of edges of any two nodes in the network. The higher the median centrality of a node, the more important it plays in connecting different nodes or node groups. The combination of these two centrality analyses can provide a more comprehensive view of understanding the structure of the network and reveal key individuals or transaction nodes in the context of a credit network.

In a sub-step S703, a cluster trend of the network is analyzed using a cluster coefficient calculation method, including calculation of local cluster coefficients and global cluster coefficients. The local cluster coefficients are implemented by calculating the connection density between neighboring nodes of the target node. Specifically, the algorithm first identifies a set of neighbor nodes that are directly connected to the target node, and then calculates the ratio between the number of edges that are actually present and the number of edges that may be present between these neighbor nodes. The higher this ratio, the tighter the relationship between the neighbors of the target node. The global cluster coefficients evaluate the connectivity of the whole network from a macroscopic point of view. This is typically accomplished by calculating the ratio of the number of all triangular closed loops in the network to the number of triangular closed loops that may be present. Calculation of these cluster coefficients not only reveals the tight connection patterns within the network, but also helps to understand the overall structural characteristics of the network, such as community formation and robustness of the network.

In the sub-step S704, a key node and risk connection identification method is used in combination with the node centrality analysis report and the cluster coefficient analysis report, aiming at determining strategic nodes in the network by synthesizing centrality and betting centrality results and revealing risk links by analyzing abnormal transaction patterns among the nodes. The key to this step is to integrate the analysis results to identify nodes that are critical to the overall credit transaction network architecture. These nodes may be key individuals controlling the flow of a large number of transactions, or important nodes bridging multiple transaction groups. At the same time, by examining the transaction patterns between these nodes, particularly transaction activities that differ significantly from conventional patterns, potentially risky connections may be revealed. These risk connections may be directed to problems with fraudulent activity, credit risk concentration, etc. By the method, a comprehensive credit network structure analysis report can be generated, so that not only is the key structure characteristics of the network disclosed, but also potential risk points can be pointed out, and important basis is provided for risk management and decision making.

Referring to fig. 9, based on the analysis result of the credit network structure, the steps of analyzing the network data by adopting the machine learning technology, identifying the abnormal transaction mode and the potential credit fraud, and generating the credit risk identification report are specifically as follows:

S803: based on the feature selection and pattern recognition results, an isolated forest algorithm is applied to analyze the abnormal transaction pattern, abnormal points are isolated by randomly selecting features and randomly segmenting feature values, the abnormal pattern is detected, a support vector machine is used to analyze, a decision boundary surrounding normal data points in a feature space is constructed, the abnormal points are separated, and an abnormal transaction pattern analysis result is generated;

S804: and integrating and explaining the analysis result by adopting a data visualization and interpretation model, improving the intuitiveness of the analysis result by a graphic representation method, and generating a credit risk identification report by providing logic and basis for model decision by the interpretation model.

In step S801, data exists in the form of multidimensional network data, and a Principal Component Analysis (PCA) algorithm is used to dimensionality reduce the data. First, the algorithm calculates the covariance matrix of the data, and then finds out the eigenvalues and eigenvectors of the covariance matrix. These feature vectors are considered the principal components of the data and are ordered in terms of their contribution to the data variance. The first few principal components are selected to reduce the dimensionality of the data while retaining the most important information, namely the key variability of the data. This process effectively reduces redundant information, optimizes the data processing efficiency of subsequent steps, and generates optimized network data that will be used for the next feature selection.

Step S802 uses a decision tree algorithm for feature selection. The optimized network data is analyzed by a decision tree algorithm that screens classification features by creating a tree structure of decision rules. When constructing the decision tree, the algorithm selects the optimal attribute for node splitting to maximize the definition of the classification. A random forest algorithm is then applied to the screened features for pattern recognition, which consists of a plurality of decision trees, each training a sub-sample of the dataset. By integrating the results of all trees, the random forest improves the stability and accuracy of prediction, and generates feature selection and pattern recognition results, which provide a basis for recognizing normal and abnormal transaction patterns.

In step S803, an orphan forest algorithm is applied to the feature selection and pattern recognition results, in particular to detect abnormal transaction patterns in the data. The algorithm isolates abnormal points in the data by randomly selecting features and randomly segmenting feature values. This method is particularly suitable for detection of outliers, since outliers tend to be easily isolated. Support Vector Machine (SVM) algorithms are then used for further analysis to isolate outliers by constructing a decision boundary that maximizes normal data point spacing. This step generates an abnormal transaction pattern analysis result, providing insight into potential risks.

In step S804, the analysis results of the previous steps are integrated, and the results are integrated and interpreted using the data visualization tool and the interpretation model. The visualization of the data improves the intuitiveness of the result through the graphic representation method, so that complex data and modes are easier to understand. An explanatory model, such as a decision tree, rule list, etc., provides the logic and basis for model decisions, helping the user understand how the model makes predictions. The finally generated credit risk identification report not only shows the result in detail, but also explains the reason behind the result, thereby providing a feasible basis for credit risk management.

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims

1. The personal credit report query monitoring and early warning method based on artificial intelligence is characterized by comprising the following steps of:

based on personal historical credit data, adopting a time sequence analysis algorithm to analyze credit score trend and periodicity pattern, predicting potential risks and generating credit trend analysis results;

Based on the credit trend analysis result, a dynamic credit scoring model is adopted, and the prediction analysis of the future credit state is carried out by combining the current economic index and market data, so that the future credit prediction result is generated;

Based on the future credit prediction result, adopting a group intelligent algorithm to perform simulation analysis on the overall credit behavior of the market, and referring to the prediction of key credit events to generate a market credit behavior simulation result;

Based on the market credit behavior simulation result, analyzing the personal differentiated financial scene by adopting a simulated annealing algorithm, predicting credit scores under various situations, and generating a personal financial scene analysis result;

Based on personal financial transaction data and the personal financial scene analysis result, adopting a cluster analysis and pattern recognition method to analyze the consumption behavior and credit preference of the client and constructing a comprehensive client portrait;

based on the comprehensive customer portrait, adopting a feature extraction method, analyzing the behavior features of the user and the influence of the social network through t-distribution random neighborhood embedding, and generating a behavior feature analysis report;

Based on the behavior characteristic analysis report, a network analysis method is adopted to construct a network model of personal credit transaction, a network structure is analyzed, the network structure comprises node centrality and cluster coefficients, key nodes and potential risk connection are identified, and a credit network structure analysis result is generated;

based on the credit network structure analysis result, the network data is analyzed by adopting a machine learning technology, an abnormal transaction mode and potential credit fraud are identified, and a credit risk identification report is generated.

2. The personal credit report query monitoring and early warning method based on artificial intelligence of claim 1, wherein the credit trend analysis results comprise a change trend graph, a periodic pattern graph and a risk index list, the future credit prediction results comprise a prediction credit score graph and a risk inter-area estimation, the market credit behavior simulation results comprise a market trend prediction graph and a key event list, the personal financial scenario analysis results comprise a scenario analysis graph and a credit score prediction, the comprehensive customer representation comprises a consumption behavior pattern graph, a credit preference classification and a risk classification, the behavior feature analysis report comprises a behavior feature graph, a consumption habit analysis and a social network impact assessment, the credit network structure analysis results comprise a network topology graph, a key node identification and a risk connection analysis, and the credit risk identification report comprises a risk transaction pattern identification, a fraud prediction and a risk early warning index.

3. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of carrying out credit score trend and periodic pattern analysis and predicting potential risks based on personal historical credit data by adopting a time sequence analysis algorithm, and generating a credit trend analysis result are specifically as follows:

based on personal historical credit data, adopting an autoregressive moving average model, and referring to the existing credit score value to predict the change trend of the future score, and simultaneously evaluating the volatility and long-term trend in the data to generate basic trend prediction;

Based on the basic trend prediction, decomposing the time series data into trend components, seasonal components and residual components by adopting a seasonal decomposition time series analysis method, revealing periodic changes of credit scores, and generating seasonal change refinements;

based on the seasonal variation refinement, adopting an isolated forest algorithm, isolating data points by constructing a plurality of decision trees, identifying abnormal points and potential risks in the data, and generating abnormal point identification;

And integrating the basic trend prediction, seasonal variation refinement and abnormal point identification results, adopting an integrated analysis method, and providing multidimensional credit score trend assessment by comparing and fusing multiple analysis results to generate a credit trend analysis result.

4. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of adopting a dynamic credit scoring model based on the credit trend analysis result, combining current economic index and market data, performing predictive analysis of future credit status, and generating future credit prediction result are specifically as follows:

Based on the credit trend analysis result, adopting a linear regression model to analyze historical credit data, including calculating a linear relation between historical credit scores and future trends, and predicting future credit score changes by analyzing the relation between variables in a time sequence to generate historical data correlation analysis;

Based on the historical data correlation analysis, carrying out macro economic index analysis by adopting a vector autoregressive model, wherein the analysis comprises the steps of analyzing the influence of economic indexes of interest rate and failure rate on credit scores, and generating macro economic index analysis by predicting a combined effect through processing a plurality of time series data;

Based on the macro economic index analysis, a neural network model is adopted to conduct market trend influence analysis, the influence of market data comprising stock market indexes and consumer confidence on personal credit scores is analyzed, the relation between market changes and the credit scores is captured through the network model, and the market trend influence analysis is generated;

Based on the market trend influence analysis, comprehensive prediction analysis of the future credit state is carried out by adopting a multiple regression model, the influence of historical data, macro economy and market trend is integrated, and a future credit prediction result is provided through multivariate analysis.

5. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the step of generating the market credit behavior simulation result by adopting a group intelligent algorithm to perform simulation analysis on the market overall credit behavior and referring to the prediction of key credit events is specifically as follows:

based on the future credit prediction result, adopting a particle swarm optimization algorithm, and adjusting the individual position to capture the optimal solution by simulating collective behaviors to reflect the dynamic change of the credit behaviors under the differentiated market conditions so as to generate individual credit behavior dynamic simulation;

Based on the dynamic simulation of the credit behaviors of the individuals, adopting a genetic algorithm to simulate the credit behaviors of the whole market, optimizing the credit behavior mode of the market based on natural selection and genetic mechanism, predicting the credit variation trend of the whole market, and generating a credit behavior evolution simulation of the market;

Based on the market credit behavior evolution simulation, a system dynamics method is adopted to analyze the influence of critical events, particularly financial crisis or policy change, on the market credit behavior, and the propagation and influence of macroscopic events in the market are simulated by constructing a dynamic model of the market credit behavior to generate a critical event market influence simulation;

And integrating the individual credit behavior dynamic simulation, the market credit behavior evolution simulation and the key event market influence simulation results, adopting a group intelligent algorithm to perform the simulation analysis of the whole market credit behavior, and generating a market credit behavior simulation result by referring to the influence of the key credit event.

6. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the step of analyzing the personal differentiated financial scenario and predicting credit scores under various situations by adopting a simulated annealing algorithm based on the market credit behavior simulation result, and generating the personal financial scenario analysis result is specifically as follows:

Based on the market credit behavior simulation result, performing global optimization simulation of personal credit score by adopting a simulated annealing algorithm, capturing an optimal solution through random search to optimize the credit score of the individual under the condition of differentiated market, and generating a credit score global optimization simulation;

Based on the credit score global optimization simulation, adopting a decision tree algorithm to analyze credit score decisions of individuals under the target financial scene, and distinguishing credit score results under the differentiated financial scene by constructing a classification tree to generate credit score decision analysis;

Based on the credit score decision analysis, carrying out conditional probability analysis of personal credit scores by adopting a Bayesian network, and providing probability prediction for the credit scores in each scene by calculating the credit score probability in the differential financial scene to generate credit score probability prediction;

And comprehensively simulating the credit score global optimization, analyzing the credit score decision and predicting the credit score probability, adopting a simulated annealing algorithm to comprehensively analyze the credit scores of individuals under various financial situations, predicting the credit scores under different situations, and generating an analysis result of the personal financial situations.

7. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of analyzing the consumption behavior and credit preference of the customer by adopting a cluster analysis and pattern recognition method based on personal financial transaction data and the personal financial scene analysis result, and constructing an integrated customer portrait are specifically as follows:

Based on personal financial transaction data and the personal financial scene analysis result, grouping consumption behavior data by adopting a K-means clustering algorithm, and dividing consumption behaviors into a plurality of categories according to a distance minimization principle by calculating the distance between each data point and a clustering center to generate a consumption behavior clustering result;

Based on the consumption behavior clustering result, key features are extracted from consumption behavior data by adopting principal component analysis, the data is converted into a group of linear independent variables through orthogonal transformation, the data dimension is reduced, key information is highlighted, and key consumption feature extraction is generated;

based on the key consumption feature extraction, classifying and pattern recognition are carried out on the extracted features by adopting a support vector machine, differentiated consumption patterns are distinguished by capturing an optimal hyperplane, and consumption pattern classification analysis is generated;

And integrating the consumption behavior clustering result, the key consumption characteristic extraction and the consumption pattern classification analysis result, and carrying out multidimensional analysis on the consumption behavior and the credit preference of the client by adopting a pattern recognition method to construct an integrated client portrait.

8. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of analyzing the behavioral characteristics and social network influence of the user and generating a behavioral characteristic analysis report by t-distributed random neighborhood embedding based on the comprehensive customer representation are specifically as follows:

Based on the comprehensive customer portrait, carrying out mathematical transformation on the features in the original data set by adopting a principal component analysis algorithm, projecting the data into a new coordinate system, identifying key variables in the data, and extracting feature vectors and feature values of the covariance matrix by calculating the covariance matrix of the data to generate a key user feature set;

Based on the key user feature set, performing dimension reduction processing on the feature set by adopting a t-distribution random neighborhood embedding algorithm, and capturing optimal representation of similarity in a low-dimensional space through calculating similarity probability between features in a high-dimensional space, so as to reduce data dimension and generate a dimension reduction user feature map;

Based on the dimension reduction user feature map, analyzing the position of a user in a social network by adopting network centrality analysis, and generating a user social network influence map by calculating the centrality and the betweenness centrality of each node;

based on the user social network influence diagram and the dimension reduction user characteristic diagram, integrating chart information by adopting a data fusion technology, performing pattern recognition and association analysis on the integrated data by adopting a statistical analysis method, analyzing the relationship between the user behavior characteristics and the social network positions of the user, and generating a behavior characteristic analysis report.

9. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of constructing a network model of personal credit transactions, analyzing a network structure including node centrality and cluster coefficients, identifying key nodes and potential risk connections, and generating a credit network structure analysis result are specifically as follows:

Based on the behavior feature analysis report, constructing a personal credit transaction network model by adopting a social network analysis algorithm, and using a scaleless network model and a small world network model, reflecting network characteristics by calculating the connection probability and reconnection probability between nodes to generate a credit transaction network model;

based on the credit transaction network model, carrying out node centrality analysis by adopting a centrality analysis method in graph theory, identifying influence nodes in the network by calculating the direct connection number of each node, evaluating the importance of the betweenness centrality by quantifying the bridging action of the nodes in the network, and generating a node centrality analysis report;

Based on the credit transaction network model, a cluster coefficient calculation method is applied to analyze the cluster trend of the network, a local cluster coefficient method is used for identifying a tightly connected node group by calculating the connection density between adjacent nodes of a target node, and a global cluster coefficient method is used for evaluating the overall connectivity of the network to generate a cluster coefficient analysis report;

And combining the node centrality analysis report and the cluster coefficient analysis report, determining strategic nodes in the network by using a key node and risk connection identification method according to comprehensive centrality and betweenness centrality results, revealing risk links by analyzing abnormal transaction modes among the nodes, and generating a credit network structure analysis result.

10. The personal credit report query monitoring and early warning method based on artificial intelligence according to claim 1, wherein the steps of analyzing network data by machine learning technology, identifying abnormal transaction patterns and potential credit fraud and generating credit risk identification reports are specifically as follows:

Based on the credit network structure analysis result, adopting a principal component analysis algorithm to reduce the dimensionality of the network data, reducing redundant information by extracting key feature vectors of the data, and reserving key variability of the data to generate optimized network data;

Based on the optimized network data, adopting a decision tree algorithm to perform feature selection on the data, screening classification features by creating a tree structure of a decision rule, and performing pattern recognition on the screened features by using a random forest algorithm to generate feature selection and pattern recognition results;

based on the feature selection and pattern recognition result, an isolated forest algorithm is applied to analyze an abnormal transaction pattern, abnormal points are isolated by randomly selecting features and randomly segmenting feature values, the abnormal pattern is detected, a support vector machine is used to analyze, a decision boundary surrounding normal data points in a feature space is constructed, the abnormal points are separated, and an abnormal transaction pattern analysis result is generated;

And integrating and explaining the analysis result by adopting a data visualization and interpretation model according to the analysis result of the abnormal transaction mode, improving the intuitiveness of the analysis result by using a graphic representation method, and generating a credit risk identification report by providing a logic and basis of a model decision by using the interpretation model.