A personal loan risk assessment method based on big data

Abstract

The present invention discloses a personal loan risk assessment method based on big data, and specifically relates to the technical field of loan risk assessment. The method obtains original data for assessing the default risk of borrowers from multiple channels, pre-processes the obtained original data, extracts characteristic data from the original data, matches the obtained characteristic data with the characteristic data required by the personal loan credit risk indicator system formulated by the lender, outputs the matching result, effectively processes multi-source heterogeneous data, makes full use of the obtained characteristic data, improves data utilization efficiency, and classifies the failure type of the failed match when the matching result is a failed match, generates different processing strategies according to the classification result, and warns and optimizes the risk prediction model constructed by the established personal loan credit risk indicator system.

Description

Technical Field

[0001] The present invention relates to the technical field of loan risk assessment, and more specifically, to a personal loan risk assessment method based on big data. Background Art

[0002] A personal loan risk assessment method based on big data is to capture massive data and use advanced data analysis technology to conduct a comprehensive analysis of the borrower's multi-dimensional data to assess the risk of personal loans in a more accurate and comprehensive manner. Specifically, it not only relies on traditional credit history records, but also takes into account multiple data sources such as the borrower's consumption behavior, social network activities, mobile payment records, etc., and extracts the characteristic data required by the personal loan credit risk indicator system from the massive data according to the personal loan credit risk indicator system formulated by the lender, so as to build a risk prediction model and conduct risk assessment on the borrower's loan application;

[0003] The personal loan risk assessment method needs to obtain multi-dimensional data of borrowers from multiple channels. However, the difficulty of obtaining data from different channels varies. For example, there may be restrictions on technical integration and data sharing between different data sources, and laws, regulations and privacy protection policies have imposed restrictions and regulations on the collection and use of data, which will lead to differences in the dimensions of the borrower data obtained. The risk prediction model constructed by the established personal loan credit risk indicator system has a low tolerance rate for borrower data with large dimensional differences and lacks the ability to respond to emergencies.

[0004] For example, when the characteristic data of the borrower obtained is far less than the requirements of the personal loan credit risk indicator system, the risk prediction model constructed based on the personal loan credit risk indicator system formulated by the lender may cause misjudgment based on the missing characteristic data; when the characteristic data of the borrower obtained is far more than the requirements of the personal loan credit risk indicator system, the risk prediction model constructed based on the personal loan credit risk indicator system formulated by the lender will usually eliminate characteristic data other than the characteristic data required by the personal loan credit risk indicator system, and these eliminated characteristic data may have a high degree of fit with the correlation with the demand characteristic data and the importance of the prediction results, while the risk prediction model constructed by the established personal loan credit risk indicator system cannot optimize the characteristic data with high correlation with the demand characteristic data and the importance of the prediction results into the risk prediction model, resulting in reduced fault tolerance of the risk prediction model, which may cause misjudgment risks.

[0005] In order to solve the above defects, a technical solution is now provided. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a personal loan risk assessment method based on big data to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A personal loan risk assessment method based on big data, characterized in that it includes the following steps:

[0009] Step S1, obtaining original data for assessing the borrower's default risk from multiple channels, preprocessing the obtained original data, and extracting feature data from the original data;

[0010] Step S2, matching the acquired characteristic data with the characteristic data required by the personal loan credit risk indicator system formulated by the lender, and outputting the matching results, which include successful matches and failed matches;

[0011] Step S3, if the matching result is a successful match, the borrower's default risk is predicted using the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender; if the matching result is an invalid match, the invalid type of the invalid match is classified;

[0012] Step S4, generating different processing strategies according to the classification results, and providing early warning and optimization for the risk prediction model constructed by the established personal loan credit risk indicator system.

[0013] In a preferred embodiment, in step S2, a standard feature data set is established based on the feature data required by the personal loan credit risk indicator system formulated by the lender, and the standard feature data set is marked as , where u represents the sequence number of the standard characteristic data required by the personal loan credit risk indicator system formulated by the lender , v is a positive integer;

[0014] The feature data of the borrower obtained in step S1 is compared item by item with the standard feature data in the standard feature data set. If each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of feature data items of the borrower is equal to the total number of standard feature data items in the standard feature data set, a matching result of successful matching is output;

[0015] The conditions for determining that the matching result is an invalid match include the following:

[0016] Judgment condition 1: If the total number of feature data items of the borrower is less than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output;

[0017] Judgment condition 2: if after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, the number of feature data items that are successfully compared is less than the total number of standard feature data in the standard feature data set and there are feature data items in the borrower's feature data items that are not successfully compared, then the matching result of invalid matching is output;

[0018] Judgment condition three: If after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of the borrower's feature data items is greater than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output.

[0019] In a preferred embodiment, in step S3, if the matching result is a successful match, the borrower's default risk is predicted using a risk prediction model constructed by the personal loan credit risk indicator system developed by the lender;

[0020] If the matching result is a failed match, the failure type of the failed match is classified as follows:

[0021] If the failure results output according to judgment condition 1 and judgment condition 2 match, the failure type is marked as a missing type;

[0022] If the failure matching result is output according to the judgment condition three, the failure type is marked as a redundant type.

[0023] In a preferred embodiment, if the failure type is marked as a missing type, an early warning signal is generated to prompt the staff to mark the current borrower for a risk of misjudgment of the prediction;

[0024] If the failure type is marked as a redundant type, a correlation analysis is performed on the redundant feature data of the borrower and the standard feature data in the standard feature data set to evaluate the correlation between the redundant feature data and the standard feature data in the standard feature data set, and an importance analysis is performed on the redundant feature data of the borrower and the prediction results to evaluate the importance of the redundant feature data of the borrower and the prediction results.

[0025] In a preferred embodiment, the redundant feature data of the borrower is subjected to correlation analysis with the standard feature data in the standard feature data set, the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated, and the redundant feature correlation coefficient is obtained. The logic for obtaining the redundant feature correlation coefficient is as follows:

[0026] Obtain the borrower's standard feature data, and use a similarity matching algorithm to match similar borrowers from the lender's database, as follows: take the borrower's standard feature data as similar features, and mark them as a similar feature set, where each element in the set represents a different similar feature of the borrower;

[0027] The borrower's similar feature set is represented as a vector. Each element in the set represents each dimension of the vector. The similar features in the set are normalized. The similarity coefficient between the borrower and the target borrower matched in the lender's database is calculated by the following formula. The expression is as follows: , where Represents the similarity coefficient between the borrower and the target borrower matched in the lender database, Represents the similarity features of the i-th dimension of the borrower’s similarity feature vector, , n is a positive integer, Represents the similarity feature of the i-th dimension of the similarity feature vector of the matched target borrower;

[0028] The similarity coefficient between the obtained borrower and each target borrower matched in the lender database is filtered through the following filtering function , select the target borrower with the highest similarity coefficient from the target borrowers, mark the target borrower with the highest similarity coefficient as a similar borrower, obtain the standard feature data of the similar borrowers, and calculate the redundant feature correlation coefficient of the borrower's redundant feature data and the standard feature data of the similar borrowers through the following formula. The expression is as follows , where represents the redundant feature correlation coefficient, represents the actual value of the jth group of redundant feature data of the borrower, , m is a positive integer, Represents the average value of the borrower's redundant feature data, and the expression is as follows , represents the actual value of the jth group of standard feature data of similar borrowers, Represents the average value of the borrower's redundant feature data, and the expression is as follows ;

[0029] Calculate the redundant feature correlation coefficient , the expression is as follows , where Represents the qth redundant feature correlation coefficient between the borrower's redundant feature data and the standard feature data of similar borrowers.

[0030] In a preferred embodiment, the redundant feature data of the borrower and the prediction result are analyzed for importance, the importance of the redundant feature data of the borrower and the prediction result is evaluated, and the redundant feature importance coefficient is obtained. The logic for obtaining the redundant feature importance coefficient is as follows:

[0031] Use the risk prediction model constructed by the personal loan credit risk indicator system developed by the lender to predict the borrower's default risk and obtain the prediction results;

[0032] The borrower's redundant feature data and prediction results are used as training sample data to build a random forest model, which consists of multiple decision trees;

[0033] Each time a node in the decision tree is split, the Gini impurity of the redundant features before and after the split node is calculated. , the calculation expression is as follows , where Indicates the probability that the sample in the split node belongs to category a;

[0034] Calculate the reduction in Gini impurity , the expression is as follows , where represents the Gini impurity of the node before splitting, and are the Gini impurities of the left and right nodes after the split, N is the number of parent node samples, and are the number of samples of left and right nodes respectively;

[0035] Count the number of splits of each redundant feature in all decision trees, and mark the number of splits as F;

[0036] Traverse each decision tree and calculate the average depth of redundant features in all decision trees , the expression is as follows , where represents the depth of redundant features in the bth decision tree, , B is a positive integer;

[0037] Calculate the redundant feature importance coefficient , the expression is as follows .

[0038] In a preferred embodiment, the redundant feature correlation coefficient and the redundant feature importance coefficient are normalized, and a redundant feature fusion model is constructed according to the normalized redundant feature correlation coefficient and the redundant feature importance coefficient to generate a redundant feature fusion index. The model is based on the following formula , where Respectively represent the preset proportional coefficients of the redundant feature correlation coefficient and the redundant feature importance coefficient, and Both are greater than 0.

[0039] In a preferred embodiment, the redundant feature fusion index is compared with a preset redundant feature fusion index threshold to determine whether to incorporate the redundant feature into the existing standard feature data category, as follows:

[0040] If the redundant feature fusion index is greater than the redundant feature fusion index threshold, the evaluated redundant features are incorporated into the existing standard feature data category;

[0041] If the redundant feature fusion index is less than or equal to the redundant feature fusion index threshold, there is no need to include the evaluated redundant features into the existing standard feature data category.

[0042] Technical effects and advantages of the present invention:

[0043] 1. The present invention obtains original data for assessing the borrower's default risk from multiple channels, pre-processes the obtained original data, extracts feature data from the original data, matches the obtained feature data with the feature data required by the personal loan credit risk indicator system formulated by the lender, outputs the matching results, effectively processes multi-source heterogeneous data, makes full use of the obtained feature data, improves data utilization efficiency, and when the matching result is a failed match, classifies the failure type of the failed match, quantifies the differences in the feature data dimensions, generates different processing strategies according to the classification results, warns and optimizes the risk prediction model constructed by the established personal loan credit risk indicator system, and improves the prediction accuracy and stability of the model.

[0044] 2. When the features acquired by the present invention are classified as redundant features, a correlation analysis is performed between the redundant feature data of the borrower and the standard feature data in the standard feature data set, the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated, an importance analysis is performed between the redundant feature data of the borrower and the prediction results, the importance of the redundant feature data of the borrower and the prediction results is evaluated, a comprehensive analysis is performed between the correlation between the redundant feature data and the standard feature data in the standard feature data set and the importance of the redundant feature data of the borrower and the prediction results, the potential value of each redundant feature in assessing the default risk of the borrower is measured, and a judgment basis for incorporating redundant features into the existing standard feature data category is given, the redundant feature data is dynamically adjusted, the fault tolerance of the risk prediction model is improved, the risk of misjudgment is reduced, and when there are differences in the dimensions of the acquired borrower data, a more accurate risk assessment is provided with higher adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings;

[0046] Figure 1 Schematic diagram of the structure of the method of the embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Example: Figure 1 The present invention provides a personal loan risk assessment method based on big data, comprising the following steps:

[0049] Step S1, obtaining original data for assessing the borrower's default risk from multiple channels, preprocessing the obtained original data, and extracting feature data from the original data;

[0050] Obtain original data for assessing borrower default risk from multiple channels, including but not limited to credit reports, bank transaction records, social media activities, e-commerce transactions, mobile payment records, etc.

[0051] It should be noted that the raw data finally obtained for assessing the borrower's default risk needs to be based on the actual difficulty of obtaining it through different channels. If the difficulty of obtaining it through a certain channel is high, it may lead to failure in obtaining the raw data, resulting in the loss of the corresponding feature data.

[0052] Preprocess the acquired raw data, including: Data cleaning: fill missing values ​​using mean filling, interpolation or machine learning methods, identify and handle outliers, identify through statistical methods (such as standard deviation method, box plot, etc.), and handle outliers according to business rules or through replacement, deletion, etc.;

[0053] Data normalization: Unify the formats of raw data from different sources, such as date format, currency unit, etc., and standardize or normalize the raw data to eliminate the impact of dimensional differences on the model;

[0054] Extract characteristic data from raw data, for example, extract historical credit scores, historical default rates, credit card usage rates and other characteristic data from raw data of credit reports; extract monthly average balances, deposit frequency, withdrawal frequency and other characteristic data from raw data of bank transaction records; extract monthly average interaction rates, number of friends and other characteristic data from raw data of social media activities; extract monthly average consumption amount, monthly shopping frequency, proportion of large-amount consumption and other characteristic data from raw data of e-commerce transactions; extract monthly average payment amount, monthly payment frequency and other characteristic data from raw data of mobile payment records;

[0055] It should be noted that the feature data finally obtained also needs to be determined according to the type of specific original data obtained;

[0056] Step S2, matching the acquired characteristic data with the characteristic data required by the personal loan credit risk indicator system formulated by the lender, and outputting the matching results, which include successful matches and failed matches;

[0057] According to the characteristic data required by the personal loan credit risk indicator system formulated by the lender, a standard characteristic data set is established and the standard characteristic data set is marked as , where u represents the sequence number of the standard characteristic data required by the personal loan credit risk indicator system formulated by the lender , v is a positive integer;

[0058] The feature data of the borrower obtained in step S1 is compared item by item with the standard feature data in the standard feature data set. If each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of feature data items of the borrower is equal to the total number of standard feature data items in the standard feature data set, a matching result of successful matching is output;

[0059] The conditions for determining that the matching result is an invalid match include the following:

[0060] Judgment condition 1: If the total number of feature data items of the borrower is less than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output;

[0061] Judgment condition 2: if after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, the number of feature data items that are successfully compared is less than the total number of standard feature data in the standard feature data set and there are feature data items in the borrower's feature data items that are not successfully compared, then the matching result of invalid matching is output;

[0062] Judgment condition three: if after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of the borrower's feature data items is greater than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output;

[0063] It should be noted that if any of the above three judgment conditions is met, the matching result of the failed match is output;

[0064] Step S3, if the matching result is a successful match, the borrower's default risk is predicted using the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender; if the matching result is an invalid match, the invalid type of the invalid match is classified;

[0065] In step S3, if the matching result is a successful match, it means that the acquired characteristic data of the borrower can meet the characteristic data required by the personal loan credit risk indicator system formulated by the lender, and the borrower's default risk is predicted using the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender;

[0066] If the matching result is a failed match, the failure type of the failed match is classified as follows:

[0067] If the invalid matching result is output according to the judgment condition 1 and the judgment condition 2, it means that the acquired characteristic data of the borrower is missing and cannot meet the requirements of the personal loan credit risk indicator system formulated by the lender, and the invalid type is marked as the missing type;

[0068] If the invalid matching result is output according to the judgment condition three, it means that the acquired feature data of the borrower contains redundant data, and these redundant feature data may have potential value in assessing the borrower's default risk, and the invalid type is marked as redundant type;

[0069] Step S4, generating different processing strategies according to the classification results, and providing early warning and optimization for the risk prediction model constructed by the established personal loan credit risk indicator system, as follows:

[0070] If the failure type is marked as missing, it means that the acquired feature data of the borrower is missing and cannot meet the requirements of the personal loan credit risk indicator system formulated by the lender. If the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender is used to predict the borrower's default risk, the risk prediction model may default the missing feature data to unfavorable data and make a misjudgment. Therefore, when the failure type is marked as missing, an early warning signal is generated to prompt the staff to mark the current borrower for misjudgment risk.

[0071] If the failure type is marked as redundant, it means that the acquired feature data of the borrower is redundant and exceeds the requirements of the personal loan credit risk indicator system established by the lender. Specifically, this means that the feature data of the borrower contains additional feature data items beyond the standard feature data set, and these additional feature data items are not used by the current risk prediction model. The redundant feature data of the borrower is correlated with the standard feature data in the standard feature data set, and the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated. The redundant feature data of the borrower is analyzed for importance with the prediction results, and the importance of the redundant feature data of the borrower with the prediction results is evaluated;

[0072] The redundant feature data of the borrower is subjected to correlation analysis with the standard feature data in the standard feature data set, the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated, and the redundant feature correlation coefficient is obtained. The logic for obtaining the redundant feature correlation coefficient is as follows:

[0073] Obtain the borrower's standard feature data, and use a similarity matching algorithm to match similar borrowers from the lender's database, as follows: take the borrower's standard feature data as similar features, and mark them as a similar feature set, where each element in the set represents a different similar feature of the borrower;

[0074] The borrower's similar feature set is represented as a vector. Each element in the set represents each dimension of the vector. The similar features in the set are normalized. The similarity coefficient between the borrower and the target borrower matched in the lender's database is calculated by the following formula. The expression is as follows: , where Represents the similarity coefficient between the borrower and the target borrower matched in the lender database, Represents the similarity features of the i-th dimension of the borrower’s similarity feature vector, , n is a positive integer, Represents the similarity feature of the i-th dimension of the similarity feature vector of the matched target borrower;

[0075] The similarity coefficient between the obtained borrower and each target borrower matched in the lender database is filtered through the following filtering function , select the target borrower with the highest similarity coefficient from the target borrowers, mark the target borrower with the highest similarity coefficient as a similar borrower, obtain the standard feature data of the similar borrowers, and calculate the redundant feature correlation coefficient of the borrower's redundant feature data and the standard feature data of the similar borrowers through the following formula. The expression is as follows , where represents the redundant feature correlation coefficient, represents the actual value of the jth group of redundant feature data of the borrower, , m is a positive integer, Represents the average value of the borrower's redundant feature data, and the expression is as follows , represents the actual value of the jth group of standard feature data of similar borrowers, Represents the average value of the borrower's redundant feature data, and the expression is as follows ;

[0076] Calculate the redundant feature correlation coefficient , the expression is as follows , where represents the qth redundant feature correlation coefficient between the redundant feature data of the borrower and the standard feature data of similar borrowers;

[0077] The importance of the redundant feature data of the borrower and the prediction results is analyzed to evaluate the importance of the redundant feature data and the prediction results, and the redundant feature importance coefficient is obtained. The logic for obtaining the redundant feature importance coefficient is as follows:

[0078] Use the risk prediction model constructed by the personal loan credit risk indicator system developed by the lender to predict the borrower's default risk and obtain the prediction results;

[0079] The borrower's redundant feature data and prediction results are used as training sample data to build a random forest model, which consists of multiple decision trees;

[0080] Each time a node in the decision tree is split, the Gini impurity of the redundant features before and after the split node is calculated. , the calculation expression is as follows , where Indicates the probability that the sample in the split node belongs to category a;

[0081] Calculate the reduction in Gini impurity , the expression is as follows , where represents the Gini impurity of the node before splitting, and are the Gini impurities of the left and right nodes after the split, N is the number of parent node samples, and are the number of samples of left and right nodes respectively;

[0082] Count the number of splits of each redundant feature in all decision trees, and mark the number of splits as F;

[0083] Traverse each decision tree and calculate the average depth of redundant features in all decision trees , the expression is as follows , where represents the depth of redundant features in the bth decision tree, , B is a positive integer;

[0084] Calculate the redundant feature importance coefficient , the expression is as follows ;

[0085] Comprehensively analyze the correlation between redundant feature data and standard feature data in the standard feature data set and the importance of the borrower's redundant feature data to the prediction results, build a redundant feature fusion model, generate a redundant feature fusion index, and measure the potential value of each redundant feature in assessing the borrower's default risk;

[0086] The redundant feature correlation coefficient and redundant feature importance coefficient are normalized, and a redundant feature fusion model is constructed based on the normalized redundant feature correlation coefficient and redundant feature importance coefficient to generate a redundant feature fusion index. The model is based on the following formula , where Respectively represent the preset proportional coefficients of the redundant feature correlation coefficient and the redundant feature importance coefficient, and All are greater than 0;

[0087] It can be seen from the above calculation expression that the larger the redundant feature correlation coefficient and the larger the redundant feature importance coefficient, the larger the redundant feature fusion index, which indicates that the redundant feature has potential value for assessing the borrower's default risk, and the greater the probability of incorporating it into the existing standard feature data category. On the contrary, the smaller the redundant feature correlation coefficient and the smaller the redundant feature importance coefficient, the smaller the redundant feature fusion index, which indicates that the redundant feature has limited potential value for assessing the borrower's default risk, and the smaller the probability of incorporating it into the existing standard feature data category.

[0088] The redundant feature fusion index is compared with the preset redundant feature fusion index threshold to determine whether to incorporate the redundant feature into the existing standard feature data category, as follows:

[0089] If the redundant feature fusion index is greater than the redundant feature fusion index threshold, it means that the redundant feature has extremely high potential value in assessing the borrower's default risk, and the assessed redundant feature is included in the existing standard feature data category;

[0090] If the redundant feature fusion index is less than or equal to the redundant feature fusion index threshold, it means that the redundant feature has low potential value for assessing the borrower's default risk, and there is no need to include the assessed redundant feature into the existing standard feature data category;

[0091] The present invention obtains original data for assessing the borrower's default risk from multiple channels, pre-processes the obtained original data, extracts feature data from the original data, matches the obtained feature data with the feature data required by the personal loan credit risk indicator system formulated by the lender, outputs the matching result, effectively processes multi-source heterogeneous data, makes full use of the obtained feature data, improves data utilization efficiency, and when the matching result is a failed match, classifies the failure type of the failed match, quantifies the difference in the feature data dimension, generates different processing strategies according to the classification result, warns and optimizes the risk prediction model constructed by the established personal loan credit risk indicator system, and improves the prediction accuracy and stability of the model.

[0092] When the features acquired by the present invention are classified as redundant features, a correlation analysis is performed on the redundant feature data of the borrower and the standard feature data in the standard feature data set, the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated, an importance analysis is performed on the redundant feature data of the borrower and the prediction results, the importance of the redundant feature data of the borrower and the prediction results is evaluated, a comprehensive analysis is performed on the correlation between the redundant feature data and the standard feature data in the standard feature data set and the importance of the redundant feature data of the borrower and the prediction results, the potential value of each redundant feature in assessing the borrower's default risk is measured, and a judgment basis for incorporating redundant features into the existing standard feature data category is given, the redundant feature data is dynamically adjusted, the fault tolerance of the risk prediction model is improved, the risk of misjudgment is reduced, and when there are differences in the dimensions of the acquired borrower data, a more accurate risk assessment is provided with higher adaptability.

[0093] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0094] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0095] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A personal loan risk assessment method based on big data, characterized by: The steps include: Step S1, obtaining original data for assessing the borrower's default risk from multiple channels, preprocessing the obtained original data, and extracting feature data from the original data; Step S2, matching the acquired characteristic data with the characteristic data required by the personal loan credit risk indicator system formulated by the lender, and outputting the matching results, which include successful matches and failed matches; Step S3, if the matching result is a successful match, the borrower's default risk is predicted using the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender; if the matching result is an invalid match, the invalid type of the invalid match is classified; Step S4, generating different processing strategies according to the classification results, and providing early warning and optimization for the risk prediction model constructed by the established personal loan credit risk indicator system; The failure type classification result includes a missing type or a redundant type; If the failure type is marked as missing, an early warning signal is generated to prompt the staff to mark the current borrower for misjudgment risk; If the failure type is marked as a redundant type, a correlation analysis is performed on the redundant feature data of the borrower and the standard feature data in the standard feature data set to evaluate the correlation between the redundant feature data and the standard feature data in the standard feature data set, and a redundant feature correlation coefficient is obtained; an importance analysis is performed on the redundant feature data of the borrower and the prediction result to evaluate the importance of the redundant feature data of the borrower and the prediction result, and a redundant feature importance coefficient is obtained; The redundant feature correlation coefficient and redundant feature importance coefficient are normalized, and a redundant feature fusion model is constructed based on the normalized redundant feature correlation coefficient and redundant feature importance coefficient to generate a redundant feature fusion index ryg. The model is based on the following formula: Where a1 and a2 represent the preset proportional coefficients of the redundant feature correlation coefficient and the redundant feature importance coefficient, respectively, and both a1 and a2 are greater than 0; The redundant feature data of the borrower is subjected to correlation analysis with the standard feature data in the standard feature data set, the correlation between the redundant feature data and the standard feature data in the standard feature data set is evaluated, and the redundant feature correlation coefficient is obtained. The logic for obtaining the redundant feature correlation coefficient is as follows: Obtain the borrower's standard feature data, and use a similarity matching algorithm to match similar borrowers from the lender's database, as follows: take the borrower's standard feature data as similar features, and mark them as a similar feature set, where each element in the set represents a different similar feature of the borrower; The borrower's similar feature set is represented as a vector. Each element in the set represents each dimension of the vector. The similar features in the set are normalized. The similarity coefficient between the borrower and the target borrower matched in the lender's database is calculated by the following formula. The expression is as follows: In the formula, similarity(x, y) represents the similarity coefficient between the borrower and the target borrower matched in the lender database, x i Represents the similarity feature of the i-th dimension of the borrower’s similarity feature vector, i = {1, 2, 3…n}, n is a positive integer, y i Represents the similarity feature of the i-th dimension of the similarity feature vector of the matched target borrower; The similarity coefficient between the obtained borrower and each target borrower matched in the lender database is filtered out from the target borrowers with the highest similarity coefficient through the following screening function MAX=max(similarity(x,y)), and the target borrower with the highest similarity coefficient is marked as a similar borrower. The standard feature data of the similar borrower is obtained, and the redundant feature correlation coefficient of the borrower's redundant feature data and the standard feature data of the similar borrower is calculated by the following formula. The expression is as follows Where R represents the redundant feature correlation coefficient, K j represents the jth group actual value of the borrower's redundant feature data, j = {1, 2, 3...m}, m is a positive integer, Represents the average value of the borrower's redundant feature data, and the expression is as follows L j represents the actual value of the jth group of standard feature data of similar borrowers, Represents the average value of the borrower's redundant feature data, and the expression is as follows Calculate the redundant feature correlation coefficient ryx, the expression is as follows Where R q represents the qth redundant feature correlation coefficient between the redundant feature data of the borrower and the standard feature data of similar borrowers; The importance of the redundant feature data of the borrower and the prediction results is analyzed to evaluate the importance of the redundant feature data and the prediction results, and the redundant feature importance coefficient is obtained. The logic for obtaining the redundant feature importance coefficient is as follows: Use the risk prediction model constructed by the personal loan credit risk indicator system developed by the lender to predict the borrower's default risk and obtain the prediction results; The borrower's redundant feature data and prediction results are used as training sample data to build a random forest model, which consists of multiple decision trees; Each time a node in the decision tree is split, the Gini impurity G of the redundant features before and after the split node is calculated. The calculation expression is as follows: Where p a Indicates the probability that the sample in the split node belongs to category a; Calculate the reduction in Gini impurity ΔG, the expression is as follows Where G pre represents the Gini impurity of the node before splitting, G left and G right are the Gini impurities of the left and right nodes after the split, N is the number of samples of the parent node, and N left and N right are the number of samples of left and right nodes respectively; Count the number of splits of each redundant feature in all decision trees, and mark the number of splits as F; Traverse each decision tree and calculate the average depth of redundant features in all decision trees The expression is as follows In the formula sd b represents the depth of redundant features in the b-th decision tree, b = {1, 2, 3…B}, B is a positive integer; Calculate the redundant feature importance coefficient ryz, the expression is as follows 2. According to claim 1, a personal loan risk assessment method based on big data is characterized by: In step S2, a standard feature data set is established based on the feature data required by the personal loan credit risk indicator system formulated by the lender, and the standard feature data set is marked as TZ = {tz u }, where u represents the sequence number of the standard characteristic data required by the personal loan credit risk indicator system formulated by the lender u={1, 2, 3…v}, and v is a positive integer; The feature data of the borrower obtained in step S1 is compared item by item with the standard feature data in the standard feature data set. If each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of feature data items of the borrower is equal to the total number of standard feature data items in the standard feature data set, a matching result of successful matching is output; The conditions for determining that the matching result is an invalid match include the following: Judgment condition 1: If the total number of feature data items of the borrower is less than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output; Judgment condition 2: if after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, the number of feature data items that are successfully compared is less than the total number of standard feature data in the standard feature data set and there are feature data items in the borrower's feature data items that are not successfully compared, then the matching result of invalid matching is output; Judgment condition three: If after comparing the borrower's feature data with the standard feature data in the standard feature data set item by item, each feature data of the borrower can find a corresponding standard feature data item in the standard feature data set and the total number of the borrower's feature data items is greater than the total number of standard feature data items in the standard feature data set, then the matching result of invalid matching is output.

3. The method for personal loan risk assessment based on big data according to claim 2, characterized in that: In step S3, if the matching result is a successful match, the borrower's default risk is predicted using the risk prediction model constructed by the personal loan credit risk indicator system formulated by the lender; If the matching result is a failed match, the failure type of the failed match is classified as follows: If the failure results output according to judgment condition 1 and judgment condition 2 match, the failure type is marked as a missing type; If the failure matching result is output according to the judgment condition three, the failure type is marked as a redundant type.

4. The method for personal loan risk assessment based on big data according to claim 1, characterized in that: The redundant feature fusion index is compared with the preset redundant feature fusion index threshold to determine whether to incorporate the redundant feature into the existing standard feature data category, as follows: If the redundant feature fusion index is greater than the redundant feature fusion index threshold, the evaluated redundant features are incorporated into the existing standard feature data category; If the redundant feature fusion index is less than or equal to the redundant feature fusion index threshold, there is no need to include the evaluated redundant features into the existing standard feature data category.

Images