CN115293889A - Credit risk prediction model training method, electronic device and readable storage medium - Google Patents

Abstract

The application discloses a credit risk prediction model training method, electronic equipment and a readable storage medium, which are applied to the technical field of financial science and technology, wherein the credit risk prediction model training method comprises the following steps: acquiring a training sample set and sample weights of training samples in the training sample set; obtaining a credit risk prediction total model according to the training sample set and the weight of each sample, wherein the credit risk prediction total model consists of a plurality of risk prediction submodels; clustering the risk prediction submodels according to the submodel prediction result of the risk prediction submodel to obtain a submodel group; amplifying a training sample set according to the sub-model prediction result and the sub-model group; and returning to the execution step: and (4) obtaining a credit risk prediction total model through iterative training according to the training sample set and the weights of all samples until the credit risk prediction total model meets a preset iteration updating ending condition, and obtaining a target credit risk prediction total model. The method and the device solve the technical problem that the prediction accuracy of the credit risk of the user is low.

Description

Credit risk prediction model training method, electronic device and readable storage medium

technical field

The present application relates to the artificial intelligence technology field of financial technology (Fintech), and in particular to a credit risk prediction model training method, electronic equipment and a readable storage medium.

Background technique

With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, artificial intelligence, etc.) The industry also has higher requirements on the credit level of users.

At present, in order to evaluate the user's credit level, usually based on the predicted credit risk of the user behavior data of multiple trained risk sub-models and the real credit risk corresponding to the user behavior data, the total risk model to be trained is trained to obtain the total risk model, Therefore, the credit risk of users can be predicted through the total risk model, but in the total risk model trained by this method, there may be a small number of risk sub-models with high model weights, and the model weights of a large number of other risk sub-models are low, which is prone to occur due to the overall risk model. Too much reliance on a small number of risk sub-models leads to low prediction accuracy of the overall risk model, so the current user credit risk prediction accuracy is low.

Contents of the invention

The main purpose of this application is to provide a credit risk prediction model training method, electronic equipment and readable storage medium, aiming to solve the technical problem of low prediction accuracy of user credit risk in the prior art.

In order to achieve the above purpose, the present application provides a credit risk prediction model training method, which is applied to credit risk prediction equipment, and the credit risk prediction model training method includes:

Obtaining the user's historical behavior data as a training sample set and the sample weight of each training sample in the training sample set;

According to the training sample set and each of the sample weights, iteratively train to obtain the overall credit risk prediction model, wherein the overall credit risk prediction model is composed of multiple risk prediction sub-models;

Obtaining the sub-model prediction results of each of the risk prediction sub-models for the training sample set, and clustering the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model group;

Optimizing each of the sample weights according to the prediction results of each of the sub-models and each of the sub-model groups to amplify the training sample set;

Return to the execution step: According to the training sample set and the weights of each sample, iteratively train to obtain the overall credit risk prediction model until the overall credit risk prediction model meets the preset iterative update end condition, and obtain the target overall credit risk prediction model.

In order to achieve the above purpose, the present application also provides a credit risk prediction device, the credit risk prediction device is applied to credit risk prediction equipment, and the credit risk prediction device includes:

An acquisition module, configured to acquire the user's historical behavior data as a training sample set and the sample weight of each training sample in the training sample set;

A training module, configured to perform iterative training according to the training sample set and each of the sample weights to obtain an overall credit risk prediction model, wherein the overall credit risk prediction model is composed of multiple risk prediction sub-models;

The clustering module is used to obtain the sub-model prediction results of each of the risk prediction sub-models for the training sample set, and cluster the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model model group;

An amplification module, configured to optimize the weight of each sample according to the prediction results of each of the sub-models and each of the sub-model groups, so as to amplify the training sample set;

The optimization module is used to return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the overall credit risk prediction model, until the overall credit risk prediction model meets the preset iterative update end condition, and obtain the target credit Overall risk prediction model.

The present application also provides an electronic device, which includes: a memory, a processor, and the program of the credit risk prediction model training method stored on the memory and operable on the processor, the credit When the program of the risk prediction model training method is executed by the processor, the steps of the above credit risk prediction model training method can be realized.

The present application also provides a computer-readable storage medium, on which a program for implementing a credit risk prediction model training method is stored. When the program of the credit risk prediction model training method is executed by a processor, the above-mentioned The steps of the credit risk prediction model training method.

The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the above-mentioned credit risk prediction model training method are realized.

The present application provides a credit risk prediction model training method, electronic equipment and readable storage medium, compared to the predicted credit risk of user behavior data based on multiple trained risk sub-models and the real credit risk corresponding to user behavior data , to train the total risk model to be trained to obtain the total risk model, the application obtains the user's historical behavior data as the training sample set and the sample weight of each training sample in the training sample set; according to the training sample set and each The sample weights are iteratively trained to obtain a total credit risk prediction model, wherein the total credit risk prediction model is composed of a plurality of risk prediction sub-models, so as to obtain the sub-models of each of the risk prediction sub-models for the training sample set Prediction results, according to the prediction results of each of the sub-models, the risk prediction sub-models are clustered to obtain at least one sub-model group, and the cluster analysis of the risk prediction sub-models is realized, so as to obtain the credit risk prediction total The model provides a sub-model group of supplementary information, and then optimizes the weight of each sample according to the prediction results of each sub-model and each sub-model group, thereby expanding the training sample set for the overall credit risk prediction model. Carry out incremental learning on the amplified training samples, and return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the overall credit risk prediction model until the overall credit risk prediction model meets the preset iterative update The end condition is to obtain the overall model of target credit risk prediction, iterate the appropriate sample weights for the overall credit risk prediction model, so that the model weights of each risk prediction sub-model obtained by training according to the training sample set after the iterative optimization of each sample weight are amplified The weight distribution is uniform, thereby improving the data supplement effect of the differentiated risk prediction sub-model, and avoiding the use of multiple risk sub-models trained based on the predicted credit risk of user behavior data and the real credit risk corresponding to user behavior data. When the total risk model is to be trained to obtain the total risk model, it is easy to have the technical defect that the total risk model is too dependent on a small number of risk sub-models, resulting in low prediction accuracy of the total risk model, thereby increasing the user's credit risk. prediction accuracy.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description serve to explain the principles of the application.

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, In other words, other drawings can also be obtained from these drawings without paying creative labor.

Fig. 1 is a schematic flow chart of the first embodiment of the credit risk prediction model training method of the present application;

Fig. 2 is a schematic flow chart of the second embodiment of the credit risk prediction model training method of the present application;

Fig. 3 is a schematic diagram of the devices involved in the credit risk prediction model training method of the present application;

FIG. 4 is a schematic diagram of the equipment structure of the hardware operating environment involved in the credit risk prediction model training method in the embodiment of the present application.

The realization of the purpose, functions and advantages of the present application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.

Detailed ways

In order to make the above objects, features and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Apparently, the described embodiments are only some of the embodiments of this application, not all of them. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

Embodiment one

The embodiment of the present application provides a credit risk prediction model training method. In the first embodiment of the credit risk prediction model training method of the present application, referring to FIG. 1, the credit risk prediction model training method includes:

Step S10, acquiring the user's historical behavior data as a training sample set and the sample weight of each training sample in the training sample set;

In this embodiment, it should be noted that the number of users can be one or multiple, and the historical behavior data is the bank-specific behavior data of the user in the last time step, and the time step It can be set as a time period for periodic extraction, and the data quantity of the historical behavior data is multiple.

Exemplarily, step S10 includes: obtaining the behavior data of the user in the previous time step, obtaining historical behavior data, and using the historical behavior data as a training sample set, wherein the time step can be one month or one year, It can also be three years; the time step can be the best judging time experience value for judging the user's credit risk behavior, or it can be set by the user's credit risk judging party; the historical behavior data can be the same user's multiple The behavior data of the time period can also be the behavior data of different users in multiple time periods, and can also be the behavior data of different users in the same time period; obtain the sample weight of each training sample in the training sample set, wherein the The sample weight can be calculated according to the sub-model prediction results of each risk prediction sub-model for each training sample, or it can be the weight experience value of each training sample with the best credit risk prediction effect, or it can be set by the user's credit risk judge.

Step S20, according to the training sample set and each sample weight, perform iterative training to obtain a credit risk prediction overall model, wherein the credit risk prediction overall model is composed of multiple risk prediction sub-models;

Step S30, obtaining the sub-model prediction results of each of the risk prediction sub-models for the training sample set, and clustering the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model group;

Step S40, according to the prediction results of each of the sub-models and each of the sub-model groups, optimize the weight of each of the samples to expand the training sample set;

Step S50, return to the execution step: according to the training sample set and the weight of each sample, iteratively train to obtain the overall credit risk prediction model until the overall credit risk prediction model meets the preset iterative update end condition, and obtain the target credit risk prediction total model.

In this embodiment, it should be noted that the risk prediction sub-model may be a scorecard model, a tree model, or a deep learning model. The sub-model group is a cluster of at least one risk prediction sub-model.

Exemplarily, steps S20 to S50 include: according to the training sample set and each sample weight, perform iterative training on the overall credit risk prediction model to be trained to obtain the overall credit risk prediction model; obtain each of the risk prediction sub-models for The sub-model prediction results of the training sample set, according to the sub-model prediction results, perform cluster analysis or similarity analysis on the sub-model results to obtain the analysis results, and according to the analysis results, the risk The prediction sub-models are clustered to obtain at least one sub-model group; according to the prediction results of each of the sub-models and each of the sub-model groups, the weights of each of the samples are optimized, and according to the weights of each of the samples, the training sample set is Perform amplification; return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the overall credit risk prediction model, if it is detected that the overall credit risk prediction model meets the preset iterative update end condition, then the The overall credit risk forecasting model is used as the target credit risk forecasting overall model, wherein, since the training sample set is amplified by optimizing the weights of each sample, various training methods are utilized in the process of continuously iteratively training the overall credit risk forecasting model. The purpose of training the overall credit risk prediction model is to use the sample set, thereby enhancing the generalization of the target credit risk prediction overall model obtained from the final training.

Wherein, in step S20, said iterative training according to said training sample set and each said sample weight to obtain a credit risk prediction model, wherein said credit risk prediction model is composed of multiple risk prediction sub-models include:

Step S21, selecting a training sample from the training sample set, and determining a sample to be predicted according to the training sample and the sample weight corresponding to the training sample;

Step S22, inputting the samples to be predicted into each of the risk prediction sub-models to obtain the output prediction results of each sub-model;

Step S23, performing weighted aggregation on the output prediction results of each of the sub-models according to the model weights corresponding to each of the risk prediction sub-models to obtain the overall model output prediction result;

Step S24, outputting prediction results according to the overall model, optimizing the weighted weights of each of the models and each of the risk prediction sub-models;

Step S25, return to the execution step: select training samples from the training sample set, and determine the samples to be predicted according to the training samples and the sample weights corresponding to the training samples, until the weighted weights of each of the models meet the preset weight conditions and Each of the sub-model parameters satisfies the preset model parameter conditions, and the overall credit risk prediction model is obtained.

In this embodiment, it should be noted that the samples to be predicted include training samples and sample weights corresponding to the training samples, and the samples to be predicted are used for iterative training of the overall credit risk prediction model.

Exemplarily, steps S21 to S25 include: randomly selecting training samples from the training sample set, or selecting training samples from the training sample set in a preset order, wherein the preset order can be generated by each sample weight , and according to the training sample and the sample weight corresponding to the training sample, a sample to be predicted is generated; the sample to be predicted is input into each of the risk prediction sub-models respectively, and each of the risk prediction sub-models is obtained for the to-be-predicted The sub-models of the prediction samples output prediction results; according to the model weighted weights corresponding to each of the risk prediction sub-models, the output prediction results of each of the sub-models are weighted and aggregated by a preset aggregation method to obtain the sub-model output characteristics, through the The overall credit risk prediction model maps the output features of the sub-models to the output prediction results of the overall model, wherein the preset aggregation method can be a mean aggregation method, a linear aggregation method, or an accumulating aggregation method, etc. Aggregation method; obtain the real label corresponding to the training sample, optimize the model weighting weight corresponding to each of the risk prediction sub-models and the risk Predict the sub-model; return to the execution step: select training samples from the training sample set, and determine the samples to be predicted according to the training samples and the sample weights corresponding to the training samples, until the weighted weights of each of the models meet the preset weight conditions And each sub-model parameter satisfies the preset model parameter condition, so as to obtain the overall credit risk prediction model.

As an example, determining that the weighted weights of each of the models meet a preset weight condition may be: if the weighted weights of each of the models meet a preset weight range, wherein the preset weight range is a preset If the weighted weights of each model are evenly distributed, it is determined that the weighted weights of each of the models meet the preset weight condition, and if there is a weighted weight of the target model in each of the weighted weights of the models that does not meet the preset weight range, it is determined that Each of the weighted weights of the models does not meet the preset weight condition, or it is judged whether there is a weighted weight of the target model greater than the first preset weight threshold or less than the second preset weight threshold among the weighted weights of each of the models, if each of the If there is a target model weight greater than the first preset weight threshold or less than the second preset weight threshold in the model weighted weights, it is determined that each of the model weights does not meet the preset weight conditions, and if each of the model weights does not If there is a target model weight greater than the first preset weight threshold or less than the second preset weight threshold, it is determined that each of the model weights satisfies the preset weight condition.

As an example, until the weighted weight of each of the models meets the preset weight condition and each of the sub-model parameters meets the preset model parameter condition, the step of obtaining the overall credit risk prediction model may be: according to the difference Degree, construct the model loss corresponding to the overall credit risk prediction model; judge whether the model loss is convergent, if convergent, then weight the model weights corresponding to each of the risk prediction sub-models and the weight of each of the risk prediction sub-models The overall credit risk prediction model is used as the overall credit risk prediction model. If it does not converge, update the overall credit risk prediction model according to the gradient calculated by the model loss, and return to the execution step: select training samples from the training sample set, And according to the training sample and the sample weight corresponding to the training sample, the sample to be predicted is determined until the calculated model loss converges.

It can be understood that the risk prediction sub-model is obtained by iterative training of the training samples and the risk prediction sub-model to be trained. In the embodiment of the present application, the risk prediction sub-model is performed multiple times through the training samples and the corresponding weights of each sample after multiple optimizations. Iterative training makes the model weight distribution of each risk prediction sub-model uniform.

Wherein, in step S40, the step of optimizing each of the sample weights according to the prediction results of each of the sub-models and each of the sub-model groups to expand the training sample set includes:

Step S41, selecting a target sub-model group that satisfies a preset condition in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition;

Step S42, adjusting the weight of each sample according to the target sub-model group and the prediction results of each sub-model, so as to expand the training sample set.

In this embodiment, it should be noted that the preset condition is a preset sub-model group condition for screening target sub-model groups that need sample weight optimization.

Exemplarily, steps S41 to S42 include: selecting a target sub-model group satisfying preset conditions in each of the sub-model groups; adjusting the training sample set according to the target sub-model group and the training sample set The sample weight of each training sample is obtained to adjust the weight to amplify the training sample set.

Wherein, in step S41, the target sub-model group that satisfies the preset condition is selected in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition include:

Step A10, obtaining the overall model prediction result of the overall credit risk prediction model for the training sample set;

Step A20, according to the prediction results of each of the sub-models and the prediction result of the overall model, generate the sub-model group correlation between each of the sub-model groups and the overall credit risk prediction model;

Step A30, selecting a target sub-model group whose correlation of the sub-model groups is smaller than a preset correlation threshold among each of the sub-model groups.

In this embodiment, it should be noted that the preset correlation threshold is a preset critical value of the sub-model group correlation between the sub-model group that needs sample weight optimization and the overall credit risk prediction model.

Exemplarily, step A10 to step A30 include: by inputting each training sample in the training sample set into the overall credit risk prediction model to obtain the overall model prediction result; obtaining each of the sub-model prediction results and the overall model prediction result The correlation between the results, using the correlation of the results as the sub-model correlation between the risk prediction sub-model and the overall credit risk prediction model; according to the correlation of each of the sub-models, generating each sub-model group and the sub-model group correlation between the overall credit risk prediction model; selecting a target sub-model group whose correlation with the sub-model group is smaller than a preset correlation threshold among each of the sub-model groups.

As an example, the step of generating the sub-model group correlation between each of the sub-model groups and the overall credit risk forecasting model according to the sub-model correlation may be as follows: the sub-model group The sum or average value of the sub-model correlations corresponding to each of the risk prediction sub-models in , is used as the sub-model group correlation between the overall credit risk prediction models.

Wherein, in step S41, the target sub-model group that satisfies the preset condition is selected in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition ,Also includes:

Step B10, obtaining the model weights of each of the risk prediction sub-models;

Step B20, according to the weighted weight of the model, determine the importance of each sub-model group to the sub-model group of the overall credit risk prediction model;

Step B30 , selecting a target sub-model group whose importance degree of the sub-model group is less than a preset importance threshold in each sub-model group.

In this embodiment, it should be noted that the preset importance threshold is a preset critical value of the importance of the sub-model group between the sub-model group that needs sample weight optimization and the overall credit risk prediction model.

Exemplarily, steps B10 to B30 include: obtaining the model weights of each of the risk prediction sub-models; using each of the model weights as a sub-model of each of the risk prediction sub-models for the overall credit risk prediction model Importance, according to the importance of each sub-model, generate the sub-model group importance of each sub-model group for the overall credit risk prediction model; select the sub-model group importance in each sub-model group The target sub-model group that is less than the preset importance threshold.

As an example, according to the importance of each of the sub-models, the step of generating the importance of the sub-model groups of each of the sub-model groups for the overall credit risk prediction model may be: The sum or average value of the sub-model importances corresponding to the risk prediction sub-models is used as the sub-model group importance among the overall credit risk prediction models.

Wherein, in step S41, the target sub-model group that satisfies the preset condition is selected in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition ,Also includes:

Step C10 , selecting a target sub-model group whose correlation of the sub-model group is less than a preset correlation threshold and whose importance degree is less than a preset importance threshold in each of the sub-model groups.

Exemplarily, step C10 includes: according to the correlation of each of the sub-model groups and the importance of each of the sub-model groups, selecting the correlation of the sub-model groups in each of the sub-model groups is less than a preset correlation threshold and The target sub-model group whose importance degree of the sub-model group is less than a preset importance threshold.

Wherein, in step S42, the step of adjusting each sample weight according to the target sub-model group and each sub-model prediction result includes:

Step D10, according to the target sub-model group and the prediction results of each of the sub-models, determine the distribution information of the prediction results of the target sub-model group;

Step D20, adjusting the weight of each sample according to the prediction result distribution information.

In this embodiment, it should be noted that the prediction result distribution information includes distribution information of sub-model prediction results of each risk prediction sub-model in the target sub-model group. The prediction result distribution information may be the data distribution information of the prediction results of each sub-model, the vector distribution information of the prediction results of each sub-model, or the proportion information of the prediction results of each sub-model.

Exemplarily, steps D10 to D20 include: according to the prediction results of each of the sub-models, determine the distribution information of the sub-model prediction results of each risk prediction sub-model in the target sub-model group, and obtain the distribution information of the prediction results; Adjusting the weight of each sample according to the distribution information of the prediction result.

Wherein, in step D20, the step of adjusting the weight of each sample according to the distribution information of the prediction result includes:

Step D21, obtaining the correct prediction proportion of each of the training samples correctly predicted by each of the risk prediction sub-models in the prediction result distribution information, wherein the prediction result distribution information includes each risk prediction sub-model for the training sample The sub-model prediction results of ;

Step D22, judging whether the correct predicted proportion is greater than a preset proportion threshold;

Step D23, if yes, lower the sample weight corresponding to each of the training samples;

Step D24, if not, increase the sample weight corresponding to each of the training samples.

In this embodiment, it should be noted that the preset percentage threshold is a preset determination that each of the training samples successfully predicted by the target sub-model group is correctly predicted by each of the risk prediction sub-models. The critical value of the proportion of correct predictions of .

As an example, step D21 to step D24 include: the prediction result distribution includes prediction proportion information of each sub-model prediction result, the prediction proportion information includes correct prediction proportion and wrong prediction proportion, and judging the Whether the correct prediction ratio is greater than the preset ratio threshold; if the correct prediction ratio is greater than the preset ratio threshold, lower the sample weight corresponding to each of the training samples; if the correct prediction ratio is not greater than the preset If the proportion threshold is set, the sample weight corresponding to each training sample is increased.

As an example, steps D21 to D24 include: the distribution of predicted results includes data distribution information of the predicted results of each sub-model, if the data distribution information satisfies a preset data range, wherein the preset distribution range The pre-set judgment training sample is the data distribution range of each sub-model prediction result successfully predicted by the target sub-model group, then lower the sample weight corresponding to each of the training samples; if the data distribution information does not meet the preset If the data range is large, the sample weight corresponding to each training sample is increased. Alternatively, the prediction result distribution includes the vector distribution information of the prediction results of each sub-model, and if the vector distribution information satisfies a preset vector range, wherein the preset vector range is that the preset judgment training sample is determined by the target sub-model The vector distribution range of each sub-model prediction result successfully predicted by the model group, then lower the sample weight corresponding to each of the training samples; if the data distribution information does not meet the preset vector range, then increase the weight corresponding to each of the training samples The sample weight of .

The embodiment of the present application provides a credit risk prediction model training method. Compared with the predicted credit risk of user behavior data based on multiple trained risk sub-models and the real credit risk corresponding to user behavior data, the total model of training risk Perform training to obtain the method of the overall risk model. The embodiment of the present application obtains the user's historical behavior data, and determines the training sample set of the overall credit risk prediction model and the sample weight of each training sample in the training sample set according to the historical behavior data. The overall risk prediction model is obtained by multiple risk prediction sub-models through model weighting, so as to collect the sub-model prediction results of each risk prediction sub-model for the training sample set, and then classify each risk prediction sub-model according to the sub-model prediction results, and obtain At least one sub-model group realizes the cluster analysis of the risk prediction sub-models, thereby obtaining a sub-model group that can provide supplementary information for the overall credit risk prediction model, and then optimizes the weight of each sample according to the training sample set and the sub-model group , so as to amplify the training sample set for the credit risk prediction model to carry out incremental learning according to the expanded training samples, and return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the credit The overall risk forecast model, until the overall credit risk forecast model satisfies the preset iterative update end condition, obtain the target credit risk forecast overall model, and iterate appropriate sample weights for the overall credit risk forecast model, so that each sample weight after iterative optimization The model weighted weight distribution of each risk prediction sub-model obtained from the training of the expanded training sample set is uniform, thereby improving the data supplement effect of the differentiated risk prediction sub-model and avoiding the use of multiple risk sub-models trained according to For the predicted credit risk of user behavior data and the real credit risk corresponding to user behavior data, when the total risk model to be trained is trained to obtain the total risk model, it is easy to cause risk due to the total risk model being too dependent on a small number of risk sub-models. The technical defect of low prediction accuracy of the overall model improves the prediction accuracy of user credit risk.

Embodiment two

Further, referring to FIG. 2 , based on the first embodiment of the present application, in another embodiment of the present application, for the same or similar content as the first embodiment above, reference may be made to the introduction above, and details will not be repeated hereafter. On this basis, wherein, in step S30, the step of clustering the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model group includes:

Step S31, obtaining the sub-model prediction result similarity between the sub-model prediction results, and using the sub-model prediction result similarity as the sub-model similarity between the risk prediction sub-models;

Step S32, clustering each of the risk prediction sub-models according to the similarity of the sub-models to obtain at least one sub-model group.

Exemplarily, steps S31 to S32 include: calculating the similarity of sub-model prediction results between the sub-model prediction results through a preset similarity algorithm, wherein the preset similarity algorithm can be Euclidean The distance algorithm can also be the Pearson correlation coefficient algorithm, or the cosine similarity algorithm. It is understandable that since the prediction results of each sub-model are the predicted credit risk of the user, the difference between the prediction results of each sub-model It is reflected in the value, and the content of the Euclidean distance algorithm is simple, and it is necessary to ensure that each index is at the same scale level; the Pearson correlation coefficient algorithm cannot calculate the data whose variance is not 0; the cosine similarity algorithm pays more attention to the vector Therefore, it is preferable to use the Euclidean distance algorithm to take into account the efficiency and accuracy of sub-model similarity acquisition. The similarity of the predicted results of the sub-models is used as the similarity of the sub-models between the risk prediction sub-models, and each risk prediction corresponding to the target sub-model similarity greater than the model similarity threshold among the similarities of the sub-models The sub-models are clustered, wherein the model similarity threshold is a critical value for judging the similarity of sub-models with higher similarity among risk prediction sub-models, and at least one sub-model group is obtained.

As an example, step S30 includes: selecting the first central model of each sub-model group in the risk prediction sub-model, which can be selected based on experience, or can be manually set; obtain risk prediction sub-models other than each of the central models The model distance between the model and each of the first central models; according to the model distance, each of the risk prediction sub-models is assigned to the sub-model group corresponding to the first central model with the smallest model distance; according to the For each of the risk prediction sub-models in the sub-model group, determine the second central model of each sub-model group, and judge whether the second central model is consistent with the first central model, if the second central model Consistent with the first central model, then use the sub-model group as the target sub-model group, if the second central model is inconsistent with the first central model, return to the execution step: in the risk prediction sub-model Select the first central model of each sub-model group until the second central model is consistent with the first central model to obtain at least one sub-model group.

Wherein, in step S10, before the step of obtaining the user's historical behavior data as the training sample set and the sample weights of each training sample in the training sample set, it also includes:

Step S11, obtaining a real label corresponding to each of the training samples;

Step S12, according to the training sample set and each of the risk prediction sub-models, generate the sub-model prediction results of each of the risk prediction sub-models for the training sample set;

Step S13, according to the real label and sub-model prediction results, determine the number of sub-models that are correctly predicted by each of the risk prediction sub-models for each of the training samples;

Step S14, generating sample weights for each of the training samples according to the number of sub-models, preset parameters and weight smoothing coefficients.

In this embodiment, it should be noted that the real label is the real credit risk of the users in each training sample.

Exemplarily, steps S11 to S14 include: obtaining the real label corresponding to each of the training samples; mapping each of the training samples to the user's credit risk through each of the risk prediction sub-models, and obtaining each of the risk prediction sub-models The prediction result of the sub-model of the model for the training sample set; judging whether the real label is consistent with the prediction result of the sub-model, and if the prediction result of the sub-model is consistent with the real label, then determine the prediction result of the sub-model The prediction of the corresponding risk prediction sub-model is correct, accumulating the number of sub-models of the risk prediction sub-model corresponding to the prediction result of the sub-model, and returning to the execution step: judging whether the real label is consistent with the prediction result of the sub-model, if the If the real label is inconsistent with the prediction result of the sub-model, it is determined that the risk prediction sub-model corresponding to the prediction result of the sub-model is wrong, and returns to the execution step to determine whether the real label is consistent with the prediction result of the sub-model, until each All the prediction results of the sub-models have been judged, and the number of sub-models predicted correctly by each of the risk prediction sub-models for each of the training samples is obtained, and each of the training samples is generated according to the number of sub-models, preset parameters and weight smoothing coefficients. The sample weight of the sample.

Optionally, the step of generating sample weights of each of the training samples according to the number of sub-models, preset parameters and weight smoothing coefficients may specifically be:

Wherein, _w is the sample weight of each of the training samples; mi is the number of sub-models predicted correctly by each of the risk prediction sub-models for each of the training samples; α is a weight smoothing coefficient; β is a preset parameter.

As an example, the step of generating the sample weights of each of the training samples may also include: obtaining the sample size of each of the training samples, generating a normal distribution random number of the sample size, and dividing the normal distribution random number assigned to each of the training samples as a sample weight of each of the training samples.

It can be understood that the sample weight of each training sample is determined according to the number of sub-models for which the training sample is correctly predicted by each of the risk prediction sub-models, and the number of sub-models is negatively correlated with the sample weight, and the smaller the number of sub-models, the This sample is supplementary sample data that is different from other samples, so by assigning a higher sample weight to the supplementary sample data, the iterative optimization steps for subsequent adjustment of sample weights are reduced to a certain extent, thereby improving the performance of the overall credit risk prediction model. training efficiency.

Embodiment three

The embodiment of the present application also provides a credit risk prediction device, the credit risk prediction device is applied to credit risk prediction equipment, referring to Figure 3, the credit risk prediction device includes:

An acquisition module, configured to acquire the user's historical behavior data as a training sample set and the sample weight of each training sample in the training sample set;

A training module, configured to perform iterative training according to the training sample set and each of the sample weights to obtain an overall credit risk prediction model, wherein the overall credit risk prediction model is composed of multiple risk prediction sub-models;

The clustering module is used to obtain the sub-model prediction results of each of the risk prediction sub-models for the training sample set, and cluster the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model model group;

An amplification module, configured to optimize the weight of each sample according to the prediction results of each of the sub-models and each of the sub-model groups, so as to amplify the training sample set;

The optimization module is used to return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the overall credit risk prediction model, until the overall credit risk prediction model meets the preset iterative update end condition, and obtain the target credit Overall risk prediction model.

Optionally, the training module is also used for:

selecting a training sample from the training sample set, and determining a sample to be predicted according to the training sample and the sample weight corresponding to the training sample;

Inputting the samples to be predicted into each of the risk prediction sub-models to obtain the output prediction results of each sub-model;

Perform weighted aggregation on the output prediction results of each of the sub-models according to the model weighting weights corresponding to each of the risk prediction sub-models to obtain the total model output prediction results;

Outputting prediction results according to the overall model, optimizing the weighted weights of each of the models and each of the risk prediction sub-models;

Return to the execution step: select training samples from the training sample set, and determine the samples to be predicted according to the training samples and the sample weights corresponding to the training samples, until the weighted weights of each of the models meet the preset weight conditions and each of the The sub-model parameters meet the preset model parameter conditions, and the overall credit risk prediction model is obtained.

Optionally, the clustering module is also used for:

Obtaining the sub-model prediction result similarity between the sub-model prediction results, using the sub-model prediction result similarity as the sub-model similarity between the risk prediction sub-models;

According to the similarity of the sub-models, each of the risk prediction sub-models is clustered to obtain at least one sub-model group.

Optionally, the amplification module is also used for:

Selecting a target sub-model group that satisfies a preset condition in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition;

Adjusting the weight of each sample according to the target sub-model group and the prediction results of each sub-model, so as to amplify the training sample set.

Optionally, the amplification module is also used for:

Obtain the overall model prediction result of the overall credit risk prediction model for the training sample set;

generating a sub-model group correlation between each of the sub-model groups and the overall credit risk prediction model according to the prediction results of each of the sub-models and the prediction result of the overall model;

A target sub-model group whose correlation of the sub-model groups is smaller than a preset correlation threshold is selected from each of the sub-model groups.

Optionally, the amplification module is also used for:

Obtaining the model weights of each of the risk prediction sub-models;

determining the importance of each of the sub-model groups to the sub-model groups of the overall credit risk prediction model according to the weighted weight of the model;

A target sub-model group whose importance degree of the sub-model group is less than a preset importance threshold is selected from each sub-model group.

Optionally, the amplification module is also used for:

In each of the sub-model groups, a target sub-model group whose correlation of the sub-model groups is less than a preset correlation threshold and whose importance degree is less than a preset importance threshold is selected.

Optionally, the amplification module is also used for:

According to the target sub-model group and the prediction results of each of the sub-models, determine the distribution information of the prediction results of the target sub-model group;

Adjusting the weight of each sample according to the distribution information of the prediction result.

Optionally, the amplification module is also used for:

Obtain the correct prediction proportion of each of the training samples correctly predicted by each of the risk prediction sub-models in the prediction result distribution information, wherein the prediction result distribution information includes the sub-models of each risk prediction sub-model for the training samples forecast result;

judging whether the proportion of the correct prediction is greater than a preset proportion threshold;

If so, lower the sample weight corresponding to each of the training samples;

If not, increase the sample weight corresponding to each of the training samples.

Optionally, before the step of acquiring the user's historical behavior data as the training sample set and the sample weights of each training sample in the training sample set, the credit risk prediction device is further used for:

Obtain a true label corresponding to each of the training samples;

According to each of the training samples and each of the risk prediction sub-models, generate a sub-model prediction result of each of the risk prediction sub-models for the training sample set;

According to the real label and sub-model prediction results, determine the number of sub-models that are correctly predicted by each of the risk prediction sub-models for each of the training samples;

Generate sample weights for each of the training samples according to the number of sub-models, preset parameters, and weight smoothing coefficients.

The credit risk prediction device provided by the present application adopts the credit risk prediction model training method in the above-mentioned embodiments, and solves the technical problem of low prediction accuracy of user credit risk. Compared with the prior art, the beneficial effect of the credit risk prediction device provided by the embodiment of the present application is the same as that of the credit risk prediction model training method provided by the above embodiment, and other technical features of the credit risk prediction device are the same as those described above The features disclosed in the methods of the embodiments are the same, and will not be repeated here.

Embodiment four

An embodiment of the present application provides an electronic device, and the electronic device includes: at least one processor; and a memory connected in communication with the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by Execution by at least one processor, so that at least one processor can execute the credit risk prediction model training method in the above-mentioned embodiments.

Referring now to FIG. 4 , it shows a schematic structural diagram of an electronic device suitable for implementing an embodiment of the present disclosure. The electronic equipment in the embodiment of the present disclosure may include but not limited to such as mobile phone, notebook computer, digital broadcast receiver, PDA (personal digital assistant), PAD (tablet computer), PMP (portable multimedia player), vehicle terminal (such as mobile terminals such as car navigation terminals) and fixed terminals such as digital TVs, desktop computers and the like. The electronic device shown in FIG. 4 is only an example, and should not limit the functions and scope of use of the embodiments of the present disclosure.

As shown in FIG. 4, an electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.), which may be stored in a read-only memory (ROM) or loaded into a random access memory (RAM) Various appropriate actions and processing are performed by the programs in the program. In RAM, various programs and data necessary for the operation of electronic equipment are also stored. The processing means, ROM, and RAM are connected to each other via a bus. Input/output (I/O) interfaces are also connected to the bus.

Typically, the following systems can be connected to the I/O interface: input devices including, for example, touch screens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; including, for example, liquid crystal displays (LCDs), speakers, vibrators output devices such as; storage devices including, for example, magnetic tapes, hard disks, etc.; and communication devices. A communication device may allow an electronic device to communicate with other devices wirelessly or by wire to exchange data. While an electronic device is shown with various systems in the figures, it should be understood that implementing or having all of the systems shown is not a requirement. More or fewer systems may alternatively be implemented or provided.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication means, or installed from a storage means, or installed from a ROM. When the computer program is executed by the processing device, the above-mentioned functions defined in the methods of the embodiments of the present disclosure are performed.

The electronic device provided by the present application adopts the credit risk prediction model training method in the above-mentioned embodiments, and solves the technical problem of low prediction accuracy of the user's credit risk. Compared with the prior art, the beneficial effect of the electronic device provided by the embodiment of the present application is the same as that of the credit risk prediction model training method provided by the above embodiment, and other technical features of the electronic device are disclosed in the method of the above embodiment The features are the same and will not be repeated here.

It should be understood that various parts of the present disclosure may be implemented in hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics may be combined in any one or more embodiments or examples in an appropriate manner.

The above is only a specific implementation of the application, but the scope of protection of the application is not limited thereto. Anyone familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the application. Should be covered within the protection scope of this application. Therefore, the protection scope of the present application should be determined by the protection scope of the claims.

Embodiment five

This embodiment provides a computer-readable storage medium, which has computer-readable program instructions stored thereon, and the computer-readable program instructions are used to implement the method for training a credit risk prediction model in the above-mentioned embodiments.

The computer-readable storage medium provided in the embodiment of the present application may be, for example, a USB flash drive, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In this embodiment, a computer-readable storage medium may be any tangible medium containing or storing a program, and the program may be used by or in combination with an instruction execution system, system or device. Program code embodied on a computer readable storage medium may be transmitted by any appropriate medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination of the foregoing.

The above-mentioned computer-readable storage medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

The above-mentioned computer-readable storage medium carries one or more programs, and when the above-mentioned one or more programs are executed by the electronic device, the electronic device: acquires the user's historical behavior data as a training sample set and each training sample in the training sample set sample weight; according to the training sample set and each of the sample weights, the iterative training obtains the overall credit risk prediction model, wherein the overall credit risk prediction model is composed of a plurality of risk prediction sub-models; each of the risk predictions is obtained For the sub-model prediction results of the training sample set, cluster the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model group; according to the prediction results of each of the sub-models and For each of the sub-model groups, optimize the weight of each sample to amplify the training sample set; return to the execution step: according to the training sample set and each sample weight, iteratively train to obtain the overall credit risk prediction model until The overall credit risk forecasting model satisfies the preset iterative update end condition, and the target credit risk forecasting overall model is obtained.

Computer program code for carrying out the operations of the present disclosure can be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, and conventional Procedural Programming Language - such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as through an Internet service provider). Internet connection).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more logical functions for implementing specified executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified functions or operations , or may be implemented by a combination of dedicated hardware and computer instructions.

The modules involved in the embodiments described in the present disclosure may be implemented by software or by hardware. Wherein, the name of the module does not constitute a limitation of the unit itself under certain circumstances.

The computer-readable storage medium provided by the present application stores computer-readable program instructions for executing the above-mentioned credit risk prediction model training method, which solves the technical problem of low prediction accuracy of user credit risk. Compared with the prior art, the beneficial effect of the computer-readable storage medium provided by the embodiment of the present application is the same as the beneficial effect of the credit risk prediction model training method provided by the above-mentioned implementation, and will not be repeated here.

Embodiment six

The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the above-mentioned credit risk prediction model training method are realized.

The computer program product provided by this application solves the technical problem of low prediction accuracy of user credit risk. Compared with the prior art, the beneficial effect of the computer program product provided by the embodiment of the present application is the same as that of the method for training the credit risk prediction model provided by the above embodiment, and will not be repeated here.

The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structures or equivalent process transformations made by using the description of the application and the accompanying drawings are directly or indirectly used in other related technical fields. , are all included in the patent processing scope of the present application in the same way.

Claims (12)

1. A credit risk prediction model training method is characterized in that, the credit risk prediction model training method comprises: Obtaining the user's historical behavior data as a training sample set and the sample weight of each training sample in the training sample set; According to the training sample set and each of the sample weights, iteratively train to obtain the overall credit risk prediction model, wherein the overall credit risk prediction model is composed of multiple risk prediction sub-models; Obtaining the sub-model prediction results of each of the risk prediction sub-models for the training sample set, and clustering the risk prediction sub-models according to the prediction results of each of the sub-models to obtain at least one sub-model group; Optimizing each of the sample weights according to the prediction results of each of the sub-models and each of the sub-model groups to amplify the training sample set; Return to the execution step: According to the training sample set and the weights of each sample, iteratively train to obtain the overall credit risk prediction model until the overall credit risk prediction model meets the preset iterative update end condition, and obtain the target overall credit risk prediction model. 2. credit risk prediction model training method as claimed in claim 1, is characterized in that, described according to described training sample set and each described sample weight, iterative training obtains credit risk prediction total model, wherein, described credit risk prediction The steps in which the overall model is composed of multiple risk prediction sub-models include: selecting a training sample from the training sample set, and determining a sample to be predicted according to the training sample and the sample weight corresponding to the training sample; Inputting the samples to be predicted into each of the risk prediction sub-models to obtain the output prediction results of each sub-model; Perform weighted aggregation on the output prediction results of each of the sub-models according to the model weighting weights corresponding to each of the risk prediction sub-models to obtain the total model output prediction results; Outputting prediction results according to the overall model, optimizing the weighted weights of each of the models and each of the risk prediction sub-models; Return to the execution step: select training samples from the training sample set, and determine the samples to be predicted according to the training samples and the sample weights corresponding to the training samples, until the weighted weights of each of the models meet the preset weight conditions and each of the The sub-model parameters meet the preset model parameter conditions, and the overall credit risk prediction model is obtained. 3. credit risk prediction model training method as claimed in claim 1, is characterized in that, described according to each described sub-model prediction result, described risk prediction sub-model is clustered, and the step of obtaining at least one sub-model group comprises : Obtaining the sub-model prediction result similarity between the sub-model prediction results, using the sub-model prediction result similarity as the sub-model similarity between the risk prediction sub-models; According to the similarity of the sub-models, each of the risk prediction sub-models is clustered to obtain at least one sub-model group. 4. credit risk prediction model training method as claimed in claim 1, is characterized in that, described according to each described sub-model prediction result and each described sub-model group, optimize each described sample weight, to amplify described training The steps for the sample set include: Selecting a target sub-model group that satisfies a preset condition in each of the sub-model groups, wherein the preset condition includes at least one of a correlation condition and an importance condition; Adjusting the weight of each sample according to the target sub-model group and the prediction results of each sub-model, so as to amplify the training sample set. 5. credit risk prediction model training method as claimed in claim 4, is characterized in that, described in each described sub-model group, selects the target sub-model group that satisfies preset condition, and wherein, described preset condition comprises correlation The step of at least one of the condition and the importance condition includes: Obtain the overall model prediction result of the overall credit risk prediction model for the training sample set; generating a sub-model group correlation between each of the sub-model groups and the overall credit risk prediction model according to the prediction results of each of the sub-models and the prediction result of the overall model; A target sub-model group whose correlation of the sub-model groups is smaller than a preset correlation threshold is selected from each of the sub-model groups. 6. credit risk prediction model training method as claimed in claim 4, is characterized in that, described in each described sub-model group, selects the target sub-model group that satisfies preset condition, and wherein, described preset condition comprises correlation The step of at least one of the condition and the importance condition, further comprising: Obtaining the model weights of each of the risk prediction sub-models; determining the importance of each of the sub-model groups to the sub-model groups of the overall credit risk prediction model according to the weighted weight of the model; A target sub-model group whose importance degree of the sub-model group is less than a preset importance threshold is selected from each sub-model group. 7. as described in any one of claim 4 to 6 credit risk forecasting model training method, it is characterized in that, described in each described sub-model group, select the target sub-model group that satisfies preset condition, wherein, the The preset condition includes at least one of a correlation condition and an importance condition, and also includes: In each of the sub-model groups, a target sub-model group whose correlation of the sub-model groups is less than a preset correlation threshold and whose importance degree is less than a preset importance threshold is selected. 8. credit risk prediction model training method as claimed in claim 4, is characterized in that, described according to described target sub-model group and each described sub-model prediction result, the step of adjusting each described sample weight comprises: According to the target sub-model group and the prediction results of each of the sub-models, determine the distribution information of the prediction results of the target sub-model group; Adjusting the weight of each sample according to the distribution information of the prediction result. 9. credit risk prediction model training method as claimed in claim 8, is characterized in that, described according to described prediction result distribution information, the step of adjusting each described sample weight comprises: Obtain the correct prediction proportion of each of the training samples correctly predicted by each of the risk prediction sub-models in the prediction result distribution information, wherein the prediction result distribution information includes the sub-models of each risk prediction sub-model for the training samples forecast result; judging whether the proportion of the correct prediction is greater than a preset proportion threshold; If so, lower the sample weight corresponding to each of the training samples; If not, increase the sample weight corresponding to each of the training samples. 10. credit risk prediction model training method as claimed in claim 1, is characterized in that, before the step of the sample weight of each training sample in described training sample set as training sample set and described training sample set, also comprises : Obtain a true label corresponding to each of the training samples; According to each of the training samples and each of the risk prediction sub-models, generate a sub-model prediction result of each of the risk prediction sub-models for the training sample set; According to the real label and sub-model prediction results, determine the number of sub-models that are correctly predicted by each of the risk prediction sub-models for each of the training samples; Generate sample weights for each of the training samples according to the number of sub-models, preset parameters, and weight smoothing coefficients. 11. An electronic device, characterized in that the electronic device comprises: at least one processor; and, a memory communicatively coupled to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions are executed by the at least one processor, so that the at least one processor can perform any one of claims 1 to 10 The steps of the credit risk prediction model training method. 12. A computer-readable storage medium, characterized in that, the computer-readable storage medium is stored with a program for implementing a credit risk prediction model training method, and the program for realizing the credit risk prediction model training method is executed by a processor to The steps of realizing the credit risk prediction model training method according to any one of claims 1 to 10.

Images