CN117172910A - Credit evaluation method and device based on EBM model, electronic equipment and storage medium - Google Patents

Abstract

A credit assessment method, a credit assessment device, an electronic device and a storage medium based on an EBM model, wherein the credit assessment method comprises the following steps: acquiring first data of an object to be predicted, wherein the first data is credit risk related data of the object to be predicted; inputting the first data into an EBM model, and acquiring a prediction result output by the EBM model, wherein the prediction result comprises a first result which comprises whether the object to be predicted breaks down after acquiring a loan; the training process of the EBM model comprises the following steps: obtaining a credit risk related dataset, each sample of the dataset comprising a plurality of features; based on the dataset, the EBM model is trained, the EBN model being the sum of all main effect shape functions and all interaction effect shape functions. The invention can more accurately predict whether the borrower can violate the constraint.

Description

Credit assessment methods, devices, electronic equipment, and storage media based on EBM models

Technical field

The present invention relates to the technical field of credit assessment, and in particular to a credit assessment method, device, electronic equipment, and storage medium based on an EBM model.

Background technique

Industries such as third-party payment platforms are gradually gaining ground. Banks promote personal credit consumption business, which provides income but also faces corresponding operating risks. Various types of personal consumption such as house mortgages, car loans, and bank cards are in urgent need of credit protection. However, because the personal credit system is still in the stage of continuous improvement, a large number of credit customers have unclear property information and unclear economic conditions, which has brought opportunities to many lenders with weak integrity awareness, and this risk is basically borne by commercial banks. The phenomenon of defaulted loans is also detrimental to the stable operation of commercial banks, which may lead to more serious financial systemic risks. Most of the current research on credit risk assessment models pursues accurate prediction performance, but ignores the interpretability of decision-making, and accuracy and interpretability are often difficult to balance. There is an accuracy-interpretability dilemma. Therefore, there is an urgent need to study the factors that lead to borrowing. The key factors of people's default and the influence status between various default factors.

The information disclosed in this Background section is merely intended to enhance an understanding of the general background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art that is already known to those skilled in the art.

Contents of the invention

In view of the problems existing in the existing technology, the present invention provides a credit assessment method, device, electronic equipment, and storage medium based on the EBM model.

The technical solution of the present invention provides a credit assessment method based on the EBM model, which method includes:

Obtain first data of the object to be predicted, where the first data is credit risk-related data of the object to be predicted;

Input the first data into the EBM model to obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a credit risk-related data set, where each sample of the data set includes multiple features;

Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is a shape function for a single feature. The interaction effect The effect shape function is a shape function for two different described characteristics.

Optionally, the predicted result also includes a second result, and the second result includes the degree of influence of the multiple features on the first result, which is used to analyze the reason why the first result is predicted.

Optionally, obtain a credit risk-related data set. Each sample of the data set includes multiple features, further including:

Obtain a credit risk-related sample set, where each sample of the sample set includes multiple items;

The multiple items are filtered through the Lasso algorithm, and the filtered items are used as the multiple features.

Optionally, obtain credit risk-related sample sets, further including:

Obtain personal credit data disclosed by banks;

Preprocess the personal credit data to form the sample set;

Each sample in the sample set includes at least two of the following items:

Current loan interest rate, company type, working years, whether you own a house, review status, loan purpose category, initial list status of the loan, borrower's number of early repayments, cumulative amount of borrower's early repayments, type of job, number of mortgage accounts , type of education, number of family members, number of months to repay the loan, whether the borrower is in default.

Optionally, the preprocessing includes at least one of the following:

Missing value processing, outlier processing, data balancing processing and data normalization processing.

Optionally, the formula of the Lasso algorithm is:

Among them, y is the vector of whether the borrower defaults, X is the matrix of the multiple items, β is the coefficient vector, N is the number of samples, and α is the regularization strength hyperparameter.

Optionally, the training process of the EBM model also includes:

After the EBM model training is completed, a global explanation is generated;

Wherein, the global explanation includes: the importance of each of the features, and/or the functional relationship between each of the features and the first result.

The technical solution of the present invention also provides a credit assessment device based on the EBM model, which device includes:

An acquisition module, configured to acquire the first data of the object to be predicted, where the first data is the credit risk-related data of the object to be predicted;

A prediction module, configured to input the first data into the EBM model, and obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will obtain a loan. will breach the contract;

Among them, the training process of the EBM model includes:

Obtain a credit risk-related data set, where each sample of the data set includes multiple features;

Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is the shape function for a single feature, and the interaction effect shape function is for two shape functions with different characteristics.

The technical solution of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned steps are implemented. The steps of any one of the credit evaluation methods based on the EBM model.

The technical solution of the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by the processor, the credit evaluation method based on the EBM model as described in any of the above is implemented. A step of.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Description of drawings

In order to more clearly illustrate the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are the drawings of the present invention. For some embodiments, those of ordinary skill in the art can also obtain other drawings based on these drawings without exerting creative efforts.

Figure 1 is a schematic flow chart of a credit assessment method based on the EBM model provided by an embodiment of the present invention;

Figure 2 is a schematic structural diagram of a credit assessment device based on the EBM model further provided by an embodiment of the present invention;

Figure 3 is a schematic diagram of data flow of a credit risk assessment and interpretation system based on the EBM model provided by an embodiment of the present invention;

Figure 4 is a ranking diagram of feature importance provided by an embodiment of the present invention;

Figure 5 is a schematic diagram of the shape function between the "current loan interest rate" feature and the target variable as well as the distribution of the values of the feature itself provided by the embodiment of the present invention;

Figure 6 is a partial explanation diagram of a certain sample in the data set according to the embodiment of the present invention;

FIG. 7 is a schematic diagram of the physical structure of an electronic device provided by the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clear, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention. , not all examples. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention.

The credit evaluation method based on the EBM model provided by the embodiments of the present application will be described in detail below with reference to the accompanying drawings through specific embodiments and application scenarios.

Figure 1 is a schematic flow chart of a credit evaluation method based on an EBM model provided by an embodiment of the present invention. As shown in Figure 1, the technical solution of the present invention provides a model based on EBM (Explainable Boosting Machine, explainable machine learning) The credit assessment method includes the following steps:

S110. Obtain the first data of the object to be predicted, where the first data is the credit risk-related data of the object to be predicted;

S120. Input the first data into the EBM model, and obtain the prediction result output by the EBM model. The prediction result includes the first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a data set related to credit risk. Each sample of the data set includes multiple features;

Based on the data set, train the EBM model. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is the shape function for a single feature, and the interaction effect shape function is the shape function for two different features. .

The above-mentioned EBM model is used to improve the accuracy of loan default prediction for the target of prediction (borrower).

Optionally, the prediction result also includes a second result. The second result includes the degree of influence of multiple features on the first result, which is used to analyze the reasons for predicting the first result. Based on this, the target to be predicted or the approving personnel can be informed of the main reasons why they cannot obtain a loan.

Optionally, obtain a credit risk-related data set. Each sample of the data set includes multiple features, further including:

Obtain a sample set related to credit risk. Each sample of the sample set includes multiple items;

Multiple items are filtered through the Lasso (Least absolute shrinkage and selection operator) algorithm, and the filtered items are used as multiple features.

Through the above-mentioned item screening, you can control the features used to affect the results, thereby controlling the calculation amount of the entire algorithm and the items that specifically affect the prediction results.

Optionally, obtain credit risk-related sample sets, further including:

Obtain personal credit data disclosed by banks;

Preprocess personal credit data to form a sample set;

Among them, each sample in the sample set includes at least two of the following items:

Current loan interest rate, company type, working years, whether you own a house, review status, loan purpose category, initial list status of the loan, borrower's number of early repayments, cumulative amount of borrower's early repayments, type of job, number of mortgage accounts , type of education, number of family members, number of months to repay the loan, whether the borrower is in default. Whether the borrower defaults is the target variable, and the others are explanatory variables.

Optionally, preprocessing includes at least one of the following:

Missing value processing, outlier processing, data balancing processing and data normalization processing.

Alternatively, the formula of the Lasso algorithm is:

Among them, y is a vector of whether the borrower defaults, X is a matrix of multiple items, β is a coefficient vector, N is the number of samples, and α is a regularization strength hyperparameter.

During the fitting process, the sparsity of the terms is controlled by adjusting the regularization parameter α||β|| ₁ . The optimal regularization strength hyperparameters can be found using methods such as cross-validation or grid search.

According to the trained Lasso regression model, obtain the coefficients of all items; sort the coefficients from large to small in absolute value; determine the best threshold through prior knowledge and actual needs to retain items with coefficients greater than the threshold, so as to Filtered features.

Optionally, the training process of the EBM model also includes:

After the EBM model training is completed, a global explanation is generated;

Among them, the global explanation includes: the importance of each feature, and/or the functional relationship between each feature and the first result.

Through global interpretation, the EBM model can be continuously adjusted to make the model's focus more in line with user needs.

The credit evaluation device based on the EBM model provided by the present invention is described below. The credit evaluation device based on the EBM model described below and the credit evaluation method based on the EBM model described above can be mutually referenced. It should be noted that the devices described here include virtual devices on program running equipment such as computers and processors.

Figure 2 is a schematic structural diagram of a credit assessment device based on the EBM model further provided by an embodiment of the present invention. As shown in Figure 2, the technical solution of the present invention also provides a credit assessment device based on the EBM model. The device includes:

The acquisition module is used to obtain the first data of the object to be predicted, where the first data is the credit risk-related data of the object to be predicted;

The prediction module is used to input the first data into the EBM model and obtain the prediction results output by the EBM model. The prediction results include the first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a data set related to credit risk. Each sample of the data set includes multiple features;

Based on the data set, train the EBM model. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is the shape function for a single feature, and the interaction effect shape function is the shape function for two different features. .

This embodiment can more accurately predict whether a borrower will default.

In another embodiment, the present invention proposes a credit risk assessment and interpretation method based on the EBM model, which is used by financial institutions to predict credit risk default probability to reduce loan risks and losses, including the following steps: Step 1), from credit risk Collect and preprocess data from relevant data sources, including missing value processing, outlier processing, data balancing and data normalization processing, and then obtain a preprocessed data set; step 2), characteristics of the preprocessed data set Data, use Lasso to perform feature screening, retain the filtered features, and use the screened feature data in the preprocessed data set as a data set; step 3), input the credit risk data into the interpretable model EBM, and the EBM model is Learn and select important features in the data, as well as the interaction between the data, to generate interpretable prediction results; step 4), based on the prediction results of the EBM model, evaluate and classify the borrower's credit risk, and output the loan Evaluation indicators such as human default probability and prediction accuracy; step 5), interpret and analyze the credit risk assessment results based on the global explanation and local explanation (i.e., the second result) of the EBM model, and output the borrower's Credit risk factors, credit risk trends, credit risk recommendations and other information.

This invention first collects the real data of borrowers from bank loan business, performs feature selection based on Lasso after preprocessing the data, and then uses the EBM method to realize the monitoring and identification of credit risks, and conducts experimental verification. Experimental results show that the present invention has a good effect in the application of bank personal loan default prediction, has good prediction robustness, and provides a reasonable explanation for the key factors of the borrower's default and the influence of each factor.

Further, in step 1):

According to the information collection needs of the actual loan business of financial institutions, relevant characteristic data are collected in banks. The collected information includes explanatory variables, which mainly include three aspects: personal basic information, repayment ability and repayment willingness. The basic personal information is mainly composed of Characteristics such as age, occupation, education, etc. make up the repayment ability. The repayment ability mainly consists of the characteristics of assets, wages, and social relationships. The willingness to repay mainly assesses whether the person has corresponding default time, the number of early repayments, and the cumulative repayment amount.

Specifically, the explanatory variables are the current loan interest rate, company type, working years, whether there is a house, review status, loan purpose category, initial listing status of the loan, number of borrowers' prepayments, cumulative amount of borrowers' prepayments, Type of job, number of mortgage accounts, type of education, number of family members, number of months to repay the loan, etc.

Finally, there is a target variable used to determine whether the borrower defaults.

The set _{corresponding} to _the explanatory _variables is expressed as Indicates no breach of contract, 1 indicates a breach of contract.

In the missing value processing step, since the number of missing samples in the data set is small, accounting for only 9% of the total sample data, a large amount of data can still be maintained after directly deleting the missing sample data, and 9142 pieces of data were finally retained.

In the outlier processing step, due to recording errors during the information collection process and other reasons, obviously abnormal samples appear in the data set. In order to avoid the abnormal data from interfering with the judgment of breach of contract, the quantile distance method is used to conduct analysis on each feature data in the original data set. Filter and adjust the data into four equal parts, namely the first quartile (Q1, 25% quantile), the second quartile (Q2, 50% quantile) and the third quartile. Quantile (Q3, 75% quantile). IQR is the difference between Q3 and Q1, which represents the distribution range of the middle 50% of the data in the data set. Arrange the data sets in order from small to large, and then calculate Q1 and Q3. Q1 is the value at 25% of the data set, Q3 is the value at 75% of the data set. By calculating the difference between Q3 and Q1, the IQR is obtained. The formula is: IQR=Q3-Q1. Outliers are defined as values in the data set that are below Q1-1.5*IQR or above Q3+1.5*IQR. Data points that exceed the outlier boundary are considered outliers. Delete abnormal data.

In the data balancing processing step, after missing values and outliers are processed, there is a serious class imbalance problem in the data, which seriously affects the data fitting process of the EBM model. SMOTE (Synthetic Minority Over-samplingTechnique) is used. ) method to balance the data, randomly select a sample x _j from the nearest neighbor samples of each minority sample x _i in the original data, and then determine any sample x j on the connection line between the sample x _i and the nearest neighbor sample x _j Select a point, use it as a newly synthesized minority class sample, and then add it to the original data set to improve the data imbalance problem existing in the data set. The new data x _new obtained by data balancing processing of a small number of samples in the original data set, the specific processing formula is as follows:

x _new =x+rand(0,1)×(xx _n )

Among them, x _n represents the sample value of a randomly selected sample from the k nearest neighbor samples of the sample x found in the minority class sample;

In the normalization processing step, due to the different dimensions between different explanatory variables and the change intervals are also in different orders of magnitude, some variables are ignored, which affects the results of data analysis. Use the normalization method to replace the value of each feature data in the original data set with the normalized data of each feature data, where the normalized value of the i-th feature data of the j-th sample in the original data set The final value x′ _(ij) , the specific calculation formula is as follows:

Among them, x _ij represents the i-th feature data of the j-th sample in the original data set, x _min represents the minimum value of the i-th column feature data in the original data set, and x _max represents the i-th column feature data in the original data set. the maximum value;

Further, in step 2):

The collected original data contains explanatory variables of borrowers, many of which have features that are weak in determining whether to default. Therefore, this part of the content uses Lasso's feature selection method to select features in the original data set. On the one hand, it improves The prediction effect of the model, on the other hand, reduces the computational complexity.

The main idea of Lasso's feature selection is simple linear regression plus L1 regularization. In ordinary linear regression, the optimization goal is to minimize the squared error between the predicted value and the actual value. In Lasso, the optimization goal is to minimize the loss function, which consists of two parts: the squared error term (similar to linear regression) and the L1 regularization term. The L1 regularization term is the sum of the absolute values of the model coefficients, multiplied by a regularization parameter λ (lambda). The existence of this term constrains the coefficients of the model during the optimization process, tending to compress some coefficients to zero.

Lasso's optimization problem can be expressed in the following form: Minimize: squared error term + λ*Σ|β|. Among them, the squared error term measures the difference between the model's predicted value and the actual value, λ is the regularization parameter, and β represents the coefficient vector of the model. As λ increases, the optimization process will cause the coefficients of some features to gradually shrink to zero. This is because the regularization term penalizes the absolute value of the coefficient, making the optimization process more inclined to select fewer features and compress the coefficients of unimportant features to zero. Due to the influence of L1 regularization, the coefficients of some features will become zero, thus achieving feature selection. Features with non-zero coefficients are considered more important in explaining the target variable and are therefore retained. The mathematical expression of Lasso can be equivalent to the following formula:

The first half of the formula represents the fitting of the original objective function, and the second half represents the penalty term of the parameter, the penalty coefficient λ∈[0, +∞]. The smaller λ, the smaller the penalty of the model, and the retained characteristic variables The greater the number, the larger λ, the greater the penalty, and the more parameters with a coefficient of 0. In short, by introducing the L1 regularization term and adjusting the regularization parameter λ, Lasso can shrink the model coefficients and select features during the optimization process, thereby achieving the goal of reducing model complexity and preventing overfitting.

Further, in step 3):

EBM model training process: The EBM model is an additive model that uses a boosted tree model to fit each feature and find the interaction effects between features. The personal credit data is input into the EBM model, a single feature in the EBM model and a second-order interaction feature in the EBM model are determined, and the output result of the single feature and the output result of the second-order interaction feature are determined, And add them to get the final prediction.

Identify the second-order interaction characteristics of the EBM model, including:

The EBM model integrates the lifting tree model into the generalized additive model. The EBM model structure is:

in, is the main effect, ∑ _{i, j} f _ij (x _i , x _j ) is the second-order interaction effect, and g(E(Y|X)) represents how each feature affects the output of the model.

_{Given any} pair _{of features ( xi ,} For the main effect, first train the first part of the parameters corresponding to the main effect. After training the main effects of all features, calculate the residual, and then with the goal of reducing the residual, train the corresponding interaction f _ij (x _i , x _j ) The second part of the parameters. If there is a strong correlation between (x _i , x _j ), its interaction f _ij (x _i , x _j ) greatly reduces the residual error.

For each pair of feature binary functions f _ij (x _i , x _j ), use the FAST (Features from accelerated segment test, an algorithm for corner detection) algorithm to screen and select significant second-order interaction terms; The output result of the single feature and the output result of the second-order interactive feature are linearly added to obtain the final prediction result of the EBM model.

Further, in step 4):

In order to compare the prediction effect of EBM models in credit risk assessment, five commonly used comparison models were selected, namely decision tree, K nearest neighbor model, XGBoost, logistic regression model and random forest. Accuracy rate, recall rate, precision rate and F1 score (ie F1-score) are used as the evaluation criteria of the model.

During the credit risk assessment and classification, borrowers are divided into different credit risk categories based on the prediction results of the EBM model. At the same time, the difference between the predicted value and the real value of the EBM model is evaluated, and then the prediction performance of the EBM model is evaluated. The confusion matrix is commonly used for performance evaluation. See Table 1 below for details.

Table 1

Accuracy: Describe the proportion of all correctly judged results of the EBM model to the observed values of the data set.

Accuracy: Describe the proportion of correct predictions by the EBM model among the results that the EBM model prediction is Positive.

Sensitivity: It is described that the true classification result in the data set is Positive, and the EBM model predicts the correct proportion.

F1-score: Represents the harmonic mean of precision and recall. The maximum is 1 and the minimum is 0. The closer to 1, the better the effect. Among them, recall=TP/(TP+FN), TP represents TruePositive, that is, the number of samples that the model correctly identifies as positive examples; FN represents False Negative, that is, the model incorrectly identifies samples that are actually positive examples. The number of counterexamples identified.

Further, in step 5):

Based on the structural characteristics of the EBM model, the credit risk decision-making is analyzed from two perspectives: global explanation and local explanation. Global explanation is to explain the EBM model results based on the characteristic variables in the data set. Local explanation refers to the prediction result of each input sample, analyzing which features and interaction terms it is affected by, and the degree and direction of the influence. The shape function f _u of each feature of the EBM model, whether it is a single feature or a second-order interactive feature, calculates its weight, and then sorts it to find the single feature or interactive feature with higher importance. For the calculation of weights, the EBM model uses The formula for measuring f _u ,/> Considered as the standard deviation of f _u , the importance of the intercept term is not considered in the EBM model, so let E(f _u )=0, then/> Therefore, after regularizing the data, the standard deviation is used as a weight to measure the importance of each item. In this way, the credit risk assessment results are interpreted and analyzed, and information such as the borrower's credit risk factors, credit risk trends, and credit risk recommendations are output. It should be noted that step 4) and step 5) are in a parallel relationship, and the result of step 4) does not affect step 5).

Figure 3 is a schematic diagram of data flow of a credit risk assessment and interpretation system based on the EBM model. As shown in Figure 3, the embodiment of the present invention also provides a credit risk assessment and interpretation system based on the EBM model. , the system includes:

The data acquisition module is used to obtain data related to the bank's personal loan business and send it to the data preprocessing module. The characteristic information includes basic personal information, repayment ability and repayment willingness, mainly including the current loan interest rate, company type, Working years, whether you own a house, review status, loan purpose category, initial listing status of the loan, number of borrowers' early repayments, cumulative amount of borrowers' early repayments, type of job, number of mortgage accounts, type of education, number of family members, The number of repayment months for the loan and whether the borrower is in default;

The data preprocessing module is used to receive the loan data set sent by the data acquisition module, perform missing value processing, outlier processing, data balancing processing and data normalization processing on the characteristic data in the loan data set, and perform preprocessing The final loan data set is sent to the EBM model training module;

The feature extraction module uses Lasso to perform feature screening, retains the filtered features, and uses the filtered feature data in the preprocessed data set as the data set;

The EBM model training module is used to receive the preprocessed loan data set sent by the data preprocessing module, to train and verify the data, to automatically learn and select important features from the data, and among them The interaction between them generates interpretable prediction results;

The credit risk assessment and classification module is used to receive the prediction results generated by the EBM model training and verification module and compare them with the real results of the data set to evaluate and classify the borrower's credit risk;

The credit risk interpretation and analysis module is used to receive the prediction results generated by the EBM model training and verification module, and interpret and analyze the credit risk assessment results based on information such as global interpretation and local interpretation of the EBM model;

The user interface module is used to receive the prediction results produced by the EBM model training and verification module, the performance evaluation produced by the credit risk assessment and classification module, and the result interpretation produced by the credit risk interpretation and analysis module, and is used to display the credit risk assessment and Interpret the results and provide interactive functionality.

In order to verify the performance of this method in the credit risk assessment and interpretation system of the EBM model, an empirical experiment was conducted to collect a real loan data set from Zhongyuan Bank. The data sample size is 15,216 records of defaulted loans and non-defaulted loans.

The present invention provides a credit risk assessment and explanation method based on the EBM model. The EBM model is an interpretable machine learning model that can more accurately predict the borrower's credit risk by selecting and weighting key evidence; and Compared with other machine learning methods, it has the best recognition results in terms of accuracy, precision, recall and sensitivity; this helps reduce the risk of misjudgment and missed diagnosis, thereby optimizing loan decisions.

The present invention uses the EBM model for credit risk assessment and interpretation, which can make full use of customer credit data and improve the accuracy and reliability of credit risk assessment. At the same time, the present invention can also output the customer's credit rating based on the EBM model, determine the influencing factors of credit risk and their degree of influence, and explain why a certain borrower is classified into a specific credit risk category. This interpretability helps financial institutions better understand the basis for model decisions and provides transparency and increased trust to customers.

The present invention can also flexibly adjust the parameters and feature selection of the EBM model according to different scenarios and needs, thereby realizing personalized credit risk assessment and interpretation for different types of customers. In addition, the present invention can also realize dynamic optimization of customer credit risk assessment and interpretation by continuously updating the EBM model.

For empirical performance, five commonly used machine learning classification methods were selected as comparison methods, namely decision tree (Decision tree), K nearest neighbor model (KNN), XGBoost, logistic regression (Logistic) and random forest (Randomforest) and the present invention. The proposed methods are compared, and the judgment indicators used include accuracy, recall, sensitivity and F1-score.

In order to have an intuitive understanding of the bank loan data set, the data samples used in the loan data set sample are shown in Table 2 below:

Table 2

In Table 2, the 9 Lasso-selected explanatory variables with a high degree of influence corresponding to the borrower (respectively, current loan interest rate, company type, working experience, whether there is a house, review status, loan purpose category, initial list of loans) status, the number of prepayments by the borrower, the cumulative amount of prepayments by the borrower), y represents the target variable (y=1 means default, y=0 means no default)

The training samples and test samples are divided according to the ratio of 7:3. The performance comparison of the EBM method of the present invention and the comparison method (Decision tree, KNN, XGBoost, Logistic and Random forest) (training set: test set = 70:30) is shown in the following Table 3:

table 3

It can be seen from the classification results in Table 3 that the EBM model has achieved the best prediction effect. Compared with the other five methods, it has very good classification effects in accuracy, recall, precision and F1 score. The accuracy rate reached 86.6%, which is nearly 6% higher than that of other models. As we all know, in bank loan business, every 1% increase in accuracy can reduce losses by nearly one million.

The EBM model is a global interpretation of the data set. The feature-selected data set contains 9 feature variables. The target variable is to predict whether a loan defaults. The EBM model is a global interpretation of the feature variables. It contains two levels, one is the ranking of feature importance. , the second is the functional relationship between each feature and the target variable. Figure 4 is a ranking diagram of feature importance provided by the embodiment of the present invention. As shown in Figure 4, the global explanation can understand the impact of the EBM model on different features, assist bank loan personnel to grasp the key factors that affect decision-making, and thus provide more accurate information. Comprehensive insights.

Figure 5 is a schematic diagram of the shape function between the "current loan interest rate" feature and the target variable and the distribution of the values of the feature itself provided by the embodiment of the present invention. As shown in Figure 5, what is shown above is the "current loan interest rate" feature and the distribution of the value of the feature itself. The relationship between the target variables. Shown below is the probability distribution of the "current loan interest rate" feature in the data set. It can be seen from the figure that when the loan interest rate ranges from 0 to 0.5, the default probability gradually increases. When the loan interest rate exceeds 0.7, the default probability gradually decreases as the loan interest rate increases.

The local explanation of the EBM model is mainly reflected in the explanation of the effect of each feature in the sample on the results, that is, the score of each feature in the sample, and the degree of influence of each feature on the prediction result of the sample, Figure 6 is a partial interpretation schematic diagram of a certain sample in the data set of the embodiment of the present invention. As shown in Figure 6, the real result of the sample is 0, and the predicted result is also 0. As for why the model gives such a predicted result, it is The graphic chart gives the specific score and sorts the degree of impact from top to bottom. The features on the right side include "current loan interest rate", "working years", "review status", "whether there is a house" and "Company type" has a negative effect on the prediction results, and the prediction probability of the model is the sum of the scores of each feature. Partial explanations can help bank loan officers analyze the specific reasons why the borrower's loan has not been approved (if the prediction result is that the borrower will default, the borrower's loan has not been approved), and can be accurate to a specific condition to facilitate adjustment of the decision-making results.

The present invention is a credit risk assessment and interpretation method and system based on the EBM model, wherein the method includes: collecting and preprocessing data from credit risk-related data sources, including borrower's personal information, loan information, repayment records, Credit scoring, etc.; input data into the Explainable Boosting Machine (EBM) model and use EBM to generate highly interpretable rules; use EBM to assess and predict credit risk and interpret the output based on the prediction explanation for decision-makers to understand The basis for judging the model. Experimental results show that the present invention has good prediction performance and interpretability, and can be widely used in credit risk assessment and decision support in the financial field.

Figure 7 is a schematic diagram of the physical structure of an electronic device provided by the present invention. As shown in Figure 7, the electronic device may include: a processor (processor) 710, a communications interface (Communications Interface) 720, a memory (memory) 730 and a communication interface. Bus 740, in which the processor 710, the communication interface 720, and the memory 730 complete communication with each other through the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a credit evaluation method based on the EBM model, which method includes:

Obtain first data of the object to be predicted, where the first data is credit risk-related data of the object to be predicted;

Input the first data into the EBM model to obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a credit risk-related data set, where each sample of the data set includes multiple features;

Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is a shape function for a single feature. The interaction effect The effect shape function is a shape function for two different described characteristics.

In addition, the above-mentioned logical instructions in the memory 730 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program code. .

On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are read by a computer, When executed, the computer can execute the credit evaluation method based on the EBM model provided by each of the above methods. The methods include:

Obtain first data of the object to be predicted, where the first data is credit risk-related data of the object to be predicted;

Input the first data into the EBM model to obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a credit risk-related data set, where each sample of the data set includes multiple features;

Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is a shape function for a single feature. The interaction effect The effect shape function is a shape function for two different described characteristics.

In another aspect, the present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. The computer program is implemented when executed by a processor to execute the above-mentioned credit evaluation methods based on the EBM model, so The methods include:

Obtain first data of the object to be predicted, where the first data is credit risk-related data of the object to be predicted;

Input the first data into the EBM model to obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will default after obtaining the loan;

Among them, the training process of the EBM model includes:

Obtain a credit risk-related data set, where each sample of the data set includes multiple features;

Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is a shape function for a single feature. The interaction effect The effect shape function is a shape function for two different described characteristics.

The device embodiments described above are only illustrative. The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they may be located in One location, or it can be distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Persons of ordinary skill in the art can understand and implement the method without any creative effort.

Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the part of the above technical solution that essentially contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., including a number of instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be used Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent substitutions are made to some of the technical features; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A credit assessment method based on the EBM model, characterized in that the method includes: Obtain first data of the object to be predicted, where the first data is credit risk-related data of the object to be predicted; Input the first data into the EBM model to obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will default after obtaining the loan; Among them, the training process of the EBM model includes: Obtain a credit risk-related data set, where each sample of the data set includes multiple features; Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is a shape function for a single feature. The interaction effect The effect shape function is a shape function for two different described characteristics. 2. The credit evaluation method based on the EBM model according to claim 1, characterized in that the prediction result also includes a second result, and the second result includes the impact of the multiple features on the first result. The degree is used to analyze and predict the reasons for obtaining the first result. 3. The credit assessment method based on the EBM model according to claim 1, characterized in that a data set related to credit risk is obtained, and each sample of the data set includes multiple features, further including: Obtain a credit risk-related sample set, where each sample of the sample set includes multiple items; The multiple items are filtered through the Lasso algorithm, and the filtered items are used as the multiple features. 4. The credit assessment method based on the EBM model according to claim 3, characterized in that obtaining a sample set related to credit risk further includes: Obtain personal credit data disclosed by banks; Preprocess the personal credit data to form the sample set; Each sample in the sample set includes at least two of the following items: Current loan interest rate, company type, working years, whether you own a house, review status, loan purpose category, initial list status of the loan, borrower's number of early repayments, cumulative amount of borrower's early repayments, type of job, number of mortgage accounts , type of education, number of family members, number of months to repay the loan, whether the borrower is in default. 5. The credit evaluation method based on the EBM model according to claim 4, characterized in that the preprocessing includes at least one of the following: Missing value processing, outlier processing, data balancing processing and data normalization processing. 6. The credit evaluation method based on the EBM model according to claim 5, characterized in that the formula of the Lasso algorithm is: Among them, y is the vector of whether the borrower defaults, X is the matrix of the multiple items, β is the coefficient vector, N is the number of samples, and α is the regularization strength hyperparameter. 7. The credit evaluation method based on the EBM model according to claim 1, characterized in that the training process of the EBM model further includes: After the EBM model training is completed, a global explanation is generated; Wherein, the global explanation includes: the importance of each of the features, and/or the functional relationship between each of the features and the first result. 8. A credit evaluation device based on the EBM model, characterized in that the device includes: An acquisition module, configured to acquire the first data of the object to be predicted, where the first data is the credit risk-related data of the object to be predicted; A prediction module, configured to input the first data into the EBM model, and obtain the prediction results output by the EBM model. The prediction results include a first result, and the first result includes whether the object to be predicted will obtain a loan. will breach the contract; Among them, the training process of the EBM model includes: Obtain a credit risk-related data set, where each sample of the data set includes multiple features; Based on the data set, the EBM model is trained. The EBN model is the sum of all main effect shape functions and all interaction effect shape functions. The main effect shape function is the shape function for a single feature, and the interaction effect shape function is for two shape functions with different characteristics. 9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that when the processor executes the program, it implements claim 1 -The steps of the credit evaluation method based on the EBM model described in any one of 7. 10. A non-transitory computer-readable storage medium with a computer program stored thereon, characterized in that when the computer program is executed by a processor, the EBM model-based EBM model as described in any one of claims 1-7 is implemented. Steps in the Credit Assessment Method.