CN114663102A - Method, device and storage medium for predicting default of bond issuer based on semi-supervised model - Google Patents

Abstract

The invention relates to the technical field of computers, and discloses a method, equipment and a storage medium for predicting debt subject default based on a semi-supervised model, wherein the method comprises the following steps: s1: acquiring main data of a debt main body, wherein the main data comprises news public opinion information, industrial and commercial information, market evaluation information and external information, and constructing an index system of credit default risks of the debt main body through the main data; s2: constructing bottom layer characteristics from statistical analysis, service judgment and derivation to generate bottom layer factors; s3: establishing a semi-supervised model based on the combination of an unmarked sample weighting method and a scoring card model; s4: and judging and predicting default risks of the debt subject based on the semi-supervised model. The modeling method is based on the combination of an unlabeled sample weighting method and a score card model, the XGB classifier trained by positive samples and unlabeled samples is used for expanding the scale of the positive samples according to the risk ranking capacity, the samples with the highest risk probability are used as new positive samples, the score card model is trained, and the semi-supervised model is constructed.

Description

Method, device and storage medium for predicting default of bond issuer based on semi-supervised model

technical field

The invention relates to the field of computer technology, and in particular provides a method, device and storage medium for predicting the default of a bond issuer based on a semi-supervised model.

Background technique

The traditional default prediction methods for bond-issuing companies mainly use financial data and credit researcher scores to rate companies to obtain the default probability of companies. Because news and public opinion data is unstructured data, it cannot be directly used by computer models, and it is difficult to be used as a model. Therefore, how to automatically use news and public opinion to establish a model to predict the default of the bond issuer is a necessary problem to be solved by the existing technology.

Existing technologies often use financial data and credit researchers to score different dimensions of enterprises to rate bond-issuing companies, output default probability, and manually process public opinion data, with the participation of a large number of business personnel and the input of subjective evaluation, to formulate a large number of early warning rules. , which makes the current technology have insufficient mining and analysis, it is difficult to fully implement the assessment of credit risk level, and the efficiency is low and depends on subjective judgment.

In addition, the default of bond-issuing companies is a small probability event, so in data modeling, there are very few positive samples. How to use the existing samples to expand the proportion of positive samples is the key to solving the problem of model distortion.

SUMMARY OF THE INVENTION

In order to solve the problem of manual processing of public opinion data in the prior art and the participation and subjective evaluation of a large number of business personnel to formulate early warning rules, the present invention provides a method, equipment and storage medium for predicting the default of a bond issuer based on a semi-supervised model.

The technical scheme of the present invention is as follows:

A method for predicting the default of a bond issuer based on a semi-supervised model, comprising:

S1: Obtain the subject data of the bond issuer, the subject data includes news and public opinion information, industrial and commercial information, market evaluation information, and external information, and construct an indicator system of the credit default risk of the bond issuer through the subject data;

S2: Construct underlying features from statistical analysis, business judgment, and derivation, and generate underlying factors;

S3: Establish a semi-supervised model based on the combination of unlabeled sample weighting method and scorecard model;

S4: Judging and predicting the default risk of the issuer based on the semi-supervised model.

Further, the index system of S1 grades the bond issuer through basic qualification information, financial operation information, punishment information, equity pledge information, news and public opinion information, internal and external rating information, and risk related information.

Further, the S2 uses statistical indicators of logarithm, mean, mode and extreme value to mine potential information of bond issuer data.

Further, establishing the semi-supervised model of S3 includes the following steps:

S21: Adjust parameters with the AUC as the target through grid search, and train the XGBoost model to obtain a classifier that identifies whether the sample is marked;

S22: Use the calibration classifier for probability calibration, and use the output calibration of XGBoost as the probability of the approximate standard;

S23: Use the calibrated sample and take the union with the original negative label as the modeling target of the subsequent training scorecard;

S24: Use the balanced sample weighting to calculate the weight;

S25: Use chi-square binning to convert all features into ordinal categorical variables;

S26: Analyze the degree of association between the feature and the modeling target, and the collinearity between the features, and screen out the high-quality features that can be modeled;

S27: Manually optimize feature interpretability;

S28: Train the scorecard model after encoding the features with the weight of evidence;

S29: Manually check the scoring rules, and correct a few rules that are inconsistent with the response rate distribution results.

Further, the scorecard model of the S2 is a scorecard model based on logistic regression, which converts the distribution in each feature of the positive sample into evidence weight coding, and then combines the evidence weight and the β in the regression coefficient to generate a score, and the output data. The driving scorecard model reflects the information mined from the data and the operation logic of the model, and gives the scoring process and the proportion of single-factor scores for the subject of debt issuance.

Further, the semi-supervised model tests the discrimination ability through the KS evaluation model, KS>0.4.

Further, the range of the AUC of the S21 is AUC>0.7.

The present invention also provides a device for predicting the default of a bond issuer based on a semi-supervised model, and the device for predicting the default of a bond issuer based on the semi-supervised model includes:

a memory, a processor, a communication bus, and a semi-supervised model stored on the memory to predict the default procedure of the issuer,

The communication bus is used to realize the communication connection between the processor and the memory;

The processor is configured to execute the program for predicting the default of the bond issuer based on the semi-supervised model, so as to realize the steps of the method for predicting the default of the bond issuer based on the semi-supervised model as described in any one of the above.

The present invention also provides a computer-readable storage medium storing executable instructions, the storage medium storing a program for predicting the default of a bond issuer based on a semi-supervised model, and the program for predicting the default of a bond issuer based on a semi-supervised model is When executed by the processor, any one of the above-mentioned steps of the method for predicting subject default based on semi-supervised machine learning are implemented.

The beneficial effects of the present invention at least include:

(1) This modeling method is based on the combination of the unlabeled sample weighting method and the scorecard model, using the XGB classifier trained on positive samples and unlabeled samples to sort risks to expand the scale of positive samples, using high-risk probability The highest sample is used as a new positive sample to train the scorecard model. Because of the good interpretability of the scorecard model and the white-box training process, it is used as the evaluation model to output the final result;

(2) Using the positive sample and unlabeled sample learning methods in semi-supervised learning, the scale of positive samples is enlarged, and the original severely biased modeling samples are corrected, on the one hand, the possibility of labeled samples in unlabeled samples is squarely , on the other hand, it can better allow the model to learn the characteristics of bad samples, reducing the risk of the model fitting more noise caused by sample imbalance;

(3) This method generates a model based on the machine learning model in a data-driven manner, reduces the information loss caused by subjective intervention, makes the risk warning more objective, and more effectively captures the risk change of the subject's prior default.

Description of drawings

FIG. 1 is a flow chart of predicting the default of a bond issuer based on a semi-supervised model of the present invention.

FIG. 2 is a flow chart of the semi-supervised model of the present invention.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Combined with Figure 1 and Figure 2, a method for predicting the default of a bond issuer based on a semi-supervised model includes:

S1: Obtain the subject data of the bond issuer, the subject data includes news and public opinion information, industrial and commercial information, market evaluation information, and external information, and construct an indicator system of the credit default risk of the bond issuer through the subject data;

S2: Construct underlying features from statistical analysis, business judgment, and derivation, and generate underlying factors;

S3: Establish a semi-supervised model based on the combination of unlabeled sample weighting method and scorecard model;

S4: Judging and predicting the default risk of the issuer based on the semi-supervised model.

The main goal of the present invention's modeling is to predict subjects with high default probability through quantitative models, and to realize the pre-detection of default risks. The object of analysis is the subject with public opinion data, mainly from the perspectives of news public opinion and basic industrial and commercial information, to mine the potential laws and connections of the forecast subject in the three dimensions of basic qualifications, industrial and commercial changes, and public opinion changes; enrich the bottom layer through feature engineering index, explore the risk factors associated with default risk; establish a scorecard model based on semi-supervised learning, and evaluate the possibility of predicting the default risk of the subject.

In this regard, first, determine the target variable: define the target variable as a bond defaulting subject with advance warning and matching industrial and commercial information, and take it as a serious biased sample;

Secondly, through feature engineering, the underlying factors are generated. Among them, the factors generated from industrial and commercial information involve basic qualification evaluation, financial report information evaluation, penalty information evaluation, and equity pledge evaluation; the factors generated from early warning data involve the time, exposure, type, rating and sentiment label of bond early warning;

In order to tap the potential information of the data as much as possible, this feature engineering adopts the processing methods of statistical indicators such as logarithm, mean, mode and extreme value. From the perspective of the final model-entry indicators, the statistical indicators have significant predictive power;

Finally, the evaluation is carried out. The semi-supervised scorecard model is used in this modeling, which solves the problem of serious biased sample modeling, and the model has a good degree of discrimination.

The analysis process includes:

1. Definition of target variable

The goal of this modeling is to predict the default risk of the bond subject, so whether the subject defaults is the target variable.

The default data was accumulated from April 2014 to December 2019, with a total of 437 records involving 180 entities (the same entity can correspond to multiple default records), excluding 90 entities without the first pre-default warning data and those who could not match the industrial and commercial 4 subjects of the information, and the remaining 86 default subjects are used as modeling positive samples, accounting for 0.57% of all modeling samples (15134 subjects).

2. Data preparation

2.1 Construction of indicator system

Based on the modeling needs of realizing public opinion pre-warning of bond issuers' default events, an indicator system for analyzing the credit default risk of bond issuers is established from four perspectives: news public opinion, industrial and commercial information, market evaluation, and external information. The indicator system comprehensively rates bond issuers from 8 subdivisions including basic qualification information, financial operation information, punishment information, equity pledge information, news and public opinion information, internal and external rating information, and risk-related information. Among them, a total of 133 factors have been realized in this modeling, and the involved factors are shown in Table 1:

Table 1)

2.2 Feature Construction

Integrate the data tables of industry and commerce, news and public opinion, and rating in relational databases, and realize the processing and realization of the underlying features from three aspects: statistical analysis, business judgment, and derivative structure through feature engineering.

2.3 Data verification

Probe data quality to ensure feature calculations are correct. It can be divided into two parts: accuracy test and logic test.

2.3.1 Accuracy test

Count the missing, duplicate, field type, and abnormal conditions of each feature to provide direction for subsequent outlier processing. The specific inspection methods are as follows:

Missing: the number of missing, probed rows and missing proportions for each field;

Repetition: Fields with only one value (all repetitions), and check the ratio of unique values of each field;

Field Type: Whether the field type is consistent with the design;

Outliers: 3σ principle: beyond the range of plus or minus three standard deviations from the mean.

2.3.2 Logical test

Business-defined outliers;

Computational logic test: Through random sampling, a certain proportion of data is randomly selected, and index processing is performed by using different tools, and the calculation results are finally compared.

2.4 Data cleaning

Missing values and outliers will affect the factor's ability to predict the final model results. Through the statistical analysis of the modeling samples, the data noise in the modeling samples can be identified, and the data quality and model effect can be improved.

2.4.1 Missing value handling

According to different feature types, missing values can be divided into numerical variables and categorical variables. In the process of dealing with missing values, according to the different feature meanings, missing values can be processed into one category or filled in according to the mean, median, and mode, and fields that are seriously missing (more than 80% missing) are eliminated. For the excluded features, analyze the predictive ability of the feature and consider whether to convert it into a rule and enter the model. The specific processing method is as follows:

Numerical variables: Starting from the meaning of the feature, according to the principle of minimizing data noise, determine the missing assignment strategy (mean, median, mode, etc.) of the feature. For example: in the modeling process, assign the missing value to 99 according to the meaning of the characteristics such as the time of recent industrial and commercial changes, the time of the most recent disclosure of the annual report, the time of the most recent financial report disclosure, the time of establishment, and the time of registration. The features such as capital, registered capital, and number of employees are assigned the missing value as the mean value according to their meaning; the missing value is assigned to 0 according to their meaning, such as the number of branches, the number of legal person changes, etc.

Categorical variables: The missing itself may have corresponding business meanings. In order to preserve the information in the data as much as possible, the missing values of the categorical variables are usually grouped separately and assigned. For example, in the modeling process, the lack of features such as whether it has been profitable for two or more consecutive years, whether it has been in losses for two consecutive years or more, whether it is profitable in the last fiscal year, and whether it is a loss in the most recent fiscal year is classified as a separate category, and the assignment is - 1; Assign the missing value to 0 according to the meaning of whether the shareholder holding more than half of the shares is the person subject to execution, whether the subject is the person subject to execution, etc. (representing that no corresponding event has occurred).

2.4.2 Outlier Handling

This modeling has not yet generated factors that can define outliers by business. For features that cannot define outliers from business, they can be screened according to Raida's rule (3σ criterion). Although the processing of outliers will make the input of the model more stable and help the model to capture the characteristics of the data population, the existence of outliers may also be the real characteristics of the data, so this modeling does not use Raida's rule to handle abnormality value.

2.5 Division of training and test sets

In order to verify the accuracy and stability of the model training results and have good generalization ability, this modeling divides all samples into training set and test set with a ratio of 7:3.

3. Model building

3.1 Exploratory Data Analysis

Through the exploratory analysis of the data, the default subjects of bond issuance defined as target variables, that is, risk exposure samples with Y=1, total 86, accounting for only 0.57% of the total sample of bond issuance subjects, which is a typical unbalanced sample modeling question.

3.1.1 Unbalanced sample processing

Semi-supervised method based on PULearning (positive and unlabeled sample learning). By learning the existing labeled samples, find the data points that are most similar to the labeled samples in the unlabeled samples, and use them as the newly added labeled samples;

In the case that more bad samples cannot be obtained in this modeling, the idea based on PU-Learning is selected to solve the sample imbalance. The principle and model application of imbalanced sample processing will be described in detail below.

3.2 Semi-supervised models

Positive sample and unlabeled sample modeling (Positive Unlabeled Learning) in semi-supervised learning has a wide range of applications in imbalanced sample processing, potential target recognition, etc., including complaints, credit risk exposure and other scenarios with less negative labels. From the perspective of PositiveUnlabeledLearning, the known defaulting entities of bonds are part of the total number of entities with high default risk, and there are still companies with high default risk among other bond-issuing entities. That is, the samples other than this are not pure risk-free samples, but unlabeled samples, so the default early warning model is actually modeled using positive samples (subjects with default events) and unlabeled samples (default risk not exposed) .

The core idea of this modeling method is that the probability of whether a sample is marked is proportional to the probability of whether it is a positive sample, that is, in terms of sorting ability, training with positive samples and unlabeled samples is equivalent to using positive samples and negative samples. Sample training, so you can first train the classifier with positive samples and unlabeled samples, convert the probability that each sample output by the classifier is labeled into the weight of each sample, and use the sample weight to train the classifier again to get that the sample belongs to the positive sample. the probability of (modeling target).

This modeling combines the above method with the scorecard model, uses the positive sample and unlabeled sample classifier's ability to sort risks to expand the scale of positive samples, and uses the 5% samples with the highest high risk probability as new positive samples, training Scorecard model. This solves the problems of indistinguishable random and truly effective features, weak model stability and ductility caused by too few high-risk labels.

The scorecard model for this modeling is based on the following principles and methodology: select a scorecard model based on logistic regression, convert the distribution of each feature of the positive sample into evidence weight coding, and then combine the evidence weight and the β in the regression coefficient to generate a score, The output data-driven scorecard can intuitively reflect the information mined from the data and the operational logic of the model, and can clearly give the debt issuer's scoring process and the proportion of single-factor scores. The formula for converting logistic regression parameters to scorecard scores is as follows:

P ₀ is the benchmark score of the scorecard, and PDO is the score that doubles the specified rate. P ₀ and PDO are two hyperparameters of the scorecard model, which are used to control the central tendency and dispersion degree of the score. This modeling takes 60 and -10;

β is the coefficient obtained by training logistic regression, intercept is the intercept obtained by training logistic regression, and n is the number of features in the model;

Calculation of constant B:

Calculation of constant A: A=P ₀ +B×ln(P ₀ );

The fixed score of the scorecard: FixedScore=A-B×intercept;

Rating for each grade:

The semi-supervised model modeling steps are as follows:

(1) A classifier for whether the training sample is marked: Train the XGBoost model, learn to classify whether a sample will be marked, and adjust the parameters through grid search with the AUC as the target. The optimal hyperparameters are shown in Table 2:

Colsample_bytree 1 Learning_rate 0.01 Max_depth 10 N_estimators 200

Table 2)

XGBoost is an optimized distributed gradient boosting library designed to be efficient, flexible and portable, it implements machine learning algorithms under the GradientBoosting framework, XGBoost provides parallel tree boosting (also known as GBDT, GBM), which can solve many Data Science Questions.

AUC (AreaUnderCurve) is defined as the area enclosed by the coordinate axis under the ROC curve. Obviously, the value of this area will not be greater than 1. Also, since the ROC curve is generally above the straight line y=x, the value range of AUC is between 0.5 and 1. The closer the AUC is to 1.0, the higher the authenticity of the detection method; when it is equal to 0.5, the authenticity is the lowest and has no application value.

(2) Probabilistic calibration: Use CalibratedClassifierCV for probability calibration, and calibrate the output of XGBoost to approximate standard probability. The hyperparameters used are method_calibrated=isotonic, cv=3, as shown in Table 3:

Method_Calibrated Isotonic CV 3

table 3)

(3) Build an expanded modeling target: use the calibrated probability to sort the Top5% samples, and take the union with the original negative label as the modeling target for the subsequent training scorecard;

(4) Unbalanced sample weighting: The model will not ignore the misclassification of positive samples because the positive samples are only 5%, and use sklearn.utils.class_weight to calculate the weight;

(5) Feature discretization: Chi-square binning is used to convert all features into ordinal categorical variables. The hyperparameters of chi-square binning are shown in Table 4:

Max_intervals 10 Min_intervals 5 Initial_intervals 100

Table 4)

(6) Predictive power and collinearity analysis: analyze the degree of association between features and modeling targets, as well as the collinearity between features, and screen high-quality features that can be modeled;

(7) Manually optimized feature interpretability: Check the high-quality features that can be entered into the model one by one, and analyze whether the frequency distribution and response rate distribution of each value can be explained in business, and whether it may be derived from random fluctuations in the data. Adjust the grouping of features and see if the validation set has the same trend as the training set. Features that cannot be explained, have a high probability of random fluctuations in the data, or that do not match the trends of the training set and the validation set cannot be included in the model;

(8) Evidence weight conversion and scoring training: The scorecard model is trained after the features are encoded by the evidence weight, and the hyperparameters are shown in Table 5:

Max_intervals 10 Min_intervals 5 Initial_intervals 100

table 5)

(9) Scorecard adjustment: manually review the scoring rules, and correct a few rules that are inconsistent with the response rate distribution results;

The English involved in the semi-supervised model modeling step refers to the parameter settings in the code.

The final semi-supervised default model entry indicators are shown in Table 6:

Table (6)

4. Model evaluation

4.1 Evaluation method

The default warning scenario of bond issuers is not a conventional classification problem involving positive and negative samples, but a semi-supervised problem of marked positive samples + unmarked samples, and the sample ratio is difficult to meet the training requirements of traditional classification models. In this modeling sample, the number of defaulting entities is too small, and the remaining entities are actually a mixture of high-risk entities that have not yet defaulted and lower-risk entities. The accuracy index will treat the entities that the model predicts as high risk but have not yet defaulted as prediction errors, but in fact, such entities are actually close to the defaulted entities in terms of risk characteristics, but the risk has not been exposed or there are other factors outside the model that cause them to not yet default. . Therefore, the traditional accuracy index is no longer suitable for the default warning scenario of the bond issuer. In this modeling, AUC is used to evaluate the overall risk ranking ability of the model, and KS is used to evaluate the model's ability to distinguish positive and negative samples.

AUC (area under the ROC curve): To test the sorting ability of the model, it is recommended that the AUC value be above 0.7. The higher the AUC, the better the classification effect of the model, and the greater the probability that default samples will be ranked ahead of non-default samples.

0.5<A under C<1, which is better than random guess and has predictive value; A under C=0.5, same as random guess, has no predictive value;

K-S statistic: To test the discriminating ability of the model, it is recommended that the KS value be above 0.40.

0.4<KS, the model distinguishing ability is good; 0.2<KS≤0.4, the model distinguishing ability is average; KS≤0.2, the model distinguishing ability is poor.

4.2 Evaluation of Semi-Supervised Models

As shown in Tables (7) and (8), the final AUC performance on the test set is 0.9617, which is close to 1; KS is 0.7779, which is greater than 0.4, indicating that the semi-supervised model has a good risk ranking ability for predicting the default probability of the issuer and ability to distinguish. At the same time, after sorting in descending order of early warning scores, the recall rate reaches 88.37% on the top 2% threshold, which also reflects the model's good predictive ability for default risk.

Evaluation indicators full sample AUC 0.9611 KS 0.8191

Table (7)

Descending order of abnormal levels Cumulative recall Top1% 74.42% Top2% 81.40% Top5% 87.21% Top10% 90.70% all 100.00%

Table (8)

The present invention provides a method for predicting the default of a bond issuer based on a semi-supervised model, including:

Take whether the issuer defaults as the target variable;

Obtain the subject data of the bond issuer's news and public opinion information, industry and commerce information, market evaluation information, and external information, and construct an indicator system for the credit default risk of the bond issuer through the subject data. 8 sub-dimensions, including equity pledge information, news and public opinion information, internal and external rating information, and risk-related information, comprehensively rate bond issuers;

Through feature engineering, the data tables of industry and commerce, news and public opinion and rating in the relational database are integrated, and the processing of the underlying features is realized from the three aspects of statistical analysis, business judgment, and derivative structure, and the underlying factors are produced. The two parts are used to test the quality of data, ensure the correctness of feature calculation, filter outliers, eliminate missing values, improve data quality, and improve prediction ability. Statistical indicators of mode and extreme value are processed;

A semi-supervised model is established based on the combination of the unlabeled sample weighting method and the scorecard model, and the positive samples and unlabeled sample classifiers can be used to sort risks to expand the size of the positive samples, and the samples with the highest high risk probability are used as new positive samples. Train a scorecard model to solve problems such as indistinguishable random and truly effective features caused by too few high-risk labels, and weak model stability and scalability. The scorecard model is a scorecard model based on logistic regression, which converts the distribution of each feature of positive samples into evidence weight codes, and then combines the evidence weights and β in the regression coefficients to generate scores, and the output data-driven scorecard model reflects the data from the data. The mined information and the operation logic of the model clearly give the scoring process and single-factor scoring ratio of the issuer;

Judging and predicting the default risk of bond issuers based on a semi-supervised model.

Using the positive samples and unlabeled samples learning methods in semi-supervised learning, the scale of positive samples has been expanded, and the original severely biased modeling samples have been revised. On the other hand, it can better allow the model to learn the characteristics of bad samples, and reduce the risk of the model fitting more noise caused by the imbalance of samples.

The method for predicting the default of a bond issuer based on a semi-supervised model provided by the present invention includes the steps of establishing a semi-supervised model:

Through grid search, the parameters are adjusted with the AUC as the target, and the XGBoost model is trained to obtain a classifier that recognizes whether the sample is marked;

Probabilistic calibration is performed using the calibration classifier, and the output calibration of XGBoost is used as the probability of the approximate standard;

Use the calibrated sample and take the union with the original negative label as the modeling target of the subsequent training scorecard;

Calculate weights using balanced sample weighting;

Use chi-square binning to convert all features into ordinal categorical variables;

Analyze the degree of association between features and modeling targets, and the collinearity between features, and screen high-quality features that can be modeled;

Manually optimize the interpretability of features, check the high-quality features that can be entered into the model one by one, analyze whether the frequency distribution and response rate distribution of each value can be explained in business, and whether it may be caused by random fluctuations in the data, and adjust the features accordingly. Group, and check whether the validation set has the same trend as the training set, and the features that cannot be explained, have a high possibility of random fluctuations in the data, or that the trends of the training set and the validation set do not match;

The scorecard model is trained by encoding the features with the weight of evidence;

Manually review the scoring rules and correct a few rules that do not match the response rate distribution results.

The method for predicting the default of a bond issuer based on a semi-supervised model provided by the present invention includes that the semi-supervised model tests the distinguishing ability through the KS evaluation model. When KS>0.4, the distinguishing ability is good.

The method for predicting the default of a bond issuer based on a semi-supervised model provided by the present invention includes testing the sorting ability of the model through AUC. The range of AUC is AUC>0.7. the greater the probability.

The present invention also provides a device for predicting the default of a bond issuer based on a semi-supervised model, and the device for predicting the default of a bond issuer based on the semi-supervised model includes:

a memory, a processor, a communication bus, and a semi-supervised model stored on the memory to predict the default procedure of the issuer,

The communication bus is used to realize the communication connection between the processor and the memory;

The processor is configured to execute the procedure of predicting the default of the bond issuer based on the semi-supervised model, so as to realize the steps of any of the above-mentioned methods for predicting the default of the bond issuer based on the semi-supervised model.

The present invention also provides a computer-readable storage medium, which stores executable instructions, and stores a program for predicting the default of a bond issuer based on a semi-supervised model. The steps of implementing any of the above-mentioned methods for predicting subject default based on semi-supervised machine learning.

The above descriptions are only the embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related technologies Fields are similarly included in the scope of patent protection of the present invention.

Claims (9)

1. A method for predicting default of a debt subject based on a semi-supervised model is characterized in that: the method comprises the following steps:

s1: acquiring main data of a debt main body, wherein the main data comprises news public opinion information, industrial and commercial information, market evaluation information and external information, and constructing an index system of credit default risks of the debt main body through the main data;

s2: constructing bottom layer characteristics from statistical analysis, service judgment and derivation to generate bottom layer factors;

s3: establishing a semi-supervised model based on the combination of an unmarked sample weighting method and a scoring card model;

s4: and judging and predicting default risks of the debt subject based on the semi-supervised model.

2. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 1, wherein: and the index system of the S1 grades the debt main body through basic qualification information, financial management information, penalty information, share right pledge information, news public opinion information, internal and external rating information and risk associated information.

3. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 1, wherein: and S2, mining potential information of the debt subject data by adopting the statistic indexes of logarithm, mean, mode and extreme value.

4. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 1, wherein: the semi-supervised model establishing of the S3 comprises the following steps:

s21: adjusting parameters by taking AUC as a target through grid search, and training an XGboost model to obtain a classifier for identifying whether a sample is marked;

s22: performing probability calibration by using a calibration classifier, and taking the output calibration of the XGboost as the probability of an approximate standard;

s23: using the calibrated sample and the original negative label as a modeling target of a subsequent training score card;

s24: calculating weights by using the balance sample weight;

s25: converting all the characteristics into ordinal classification variables by using chi-square classification boxes;

s26: analyzing the degree of association between the features and the modeling target and the colinearity between the features, and screening high-quality features which can enter the model;

s27: manually optimizing feature interpretability;

s28: training a scoring card model after the characteristic certification is subjected to weight recoding;

s29: and manually checking the scoring rules, and correcting a few rules which are inconsistent with the response rate distribution result.

5. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 1, wherein: the scoring card model of S3 is based on a scoring card model of logistic regression, the distribution in each feature of the positive sample is converted into evidence weight codes, then scoring is generated by combining the evidence weight and beta in the regression coefficient, the output data drives the scoring card model to reflect information mined from data and the operational logic of the model, and the scoring process of a debt main body and the single factor scoring ratio are given.

6. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 1, wherein: the semi-supervised model of S3 examined discriminative power by a KS evaluation model, KS > 0.4.

7. The method for predicting default of a debt subject based on semi-supervised model as recited in claim 4, wherein: the AUC of S21 ranged from AUC > 0.7.

8. An apparatus for predicting default of a debt subject based on a semi-supervised model, characterized in that: the device for predicting debt subject default based on semi-supervised model comprises:

a memory, a processor, a communication bus, and a semi-supervised model predictive debt subject default program stored on the memory,

the communication bus is used for realizing communication connection between the processor and the memory;

the processor is used for executing the semi-supervised model based default prediction program to realize the semi-supervised model based default prediction method of the debt subject as claimed in any one of claims 1 to 7.

9. A computer-readable storage medium storing executable instructions, wherein: the storage medium stores a semi-supervised model based default prediction program for a debt subject, and the semi-supervised model based default prediction program is executed by a processor to realize the steps of the semi-supervised machine learning based default prediction method for a subject according to any one of the above claims 1-7.

Images