A credit risk assessment method, device, storage medium and equipment

Abstract

The present application discloses a credit risk assessment method, apparatus, storage medium and equipment, which extracts target indicators from the credit behavior information of the customer to be tested in advance. The target indicator is a characteristic indicator that meets the preset conditions, and the preset conditions are: there is a correlation between the characteristic indicator and the newly occurring credit behavior. The target indicator of the customer to be tested is input into the prediction model to obtain the output result of the prediction model. Among them, the prediction model is based on the training set as input, and the default probability of the samples in the training set is used as the training target, and is pre-trained. The samples include pre-acquired positive samples and negative samples. The output results include the newly occurring credit risk of the customer to be tested. It can be seen that by using the scheme described in the present application, there is no need to manually consider quantitative and qualitative indicators, and the use of the pre-trained prediction model to perform new credit risk assessment can effectively improve the accuracy of new credit risk assessment.

Description

Technical Field

[0001] The present application relates to the field of big data processing, and in particular to a credit risk assessment method, device, storage medium and equipment. Background Art

[0002] In financial business, if a company's credit is not bad, interest or principal is overdue at the end of the previous year, and the company has bad credit or interest or principal is overdue for more than a certain number of days in the next year, it is defined as a newly-occurring credit risk. How to assess a customer's newly-occurring credit risk has become the focus of financial business projects.

[0003] At present, most of the existing new credit risk assessment methods use the loan materials and financial variables provided by the enterprise in combination with business experience to determine whether the enterprise meets the credit product handling policy, thereby determining the customer's new credit risk. However, the existing technical means require a large number of quantitative and qualitative indicators to be considered manually, which is highly subjective, and the accuracy of its credit risk assessment is too low. Summary of the invention

[0004] The present application provides a credit risk assessment method, apparatus, storage medium and device, the purpose of which is to improve the accuracy of credit risk assessment of newly occurring loans.

[0005] In order to achieve the above objectives, this application provides the following technical solutions:

[0006] A credit risk assessment method, comprising:

[0007] Extracting target indicators from the credit behavior information of the customer to be tested in advance; the target indicators are characteristic indicators that meet preset conditions; the preset conditions are: the characteristic indicators are associated with the newly occurring credit behavior;

[0008] The target indicators of the customer to be tested are input into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the default probability of samples in the training set as training targets; the samples include pre-acquired positive samples and negative samples; the output result includes the newly-occurred credit risk of the customer to be tested.

[0009] Optionally, the process of obtaining the positive sample and the negative sample includes:

[0010] Extracting characteristic indicators from the credit behavior information of sample customers in advance; wherein the characteristic indicators include credit asset classification, credit overdue, and interest arrears status;

[0011] When it is detected that the level of the credit asset classification changes from a higher level to a lower level during the preset observation period, the characteristic indicator of the sample customer is marked as a positive sample;

[0012] When it is detected that the level of the credit asset classification has not changed during the preset observation period, the probability of the sample customer's transition between the dishonest state and the trustworthy state is calculated through the probability transition matrix; wherein, if the sample customer has dishonest behavior, the sample customer is determined to be in the dishonest state, otherwise, the sample customer is determined to be in the trustworthy state, and the dishonest state may be that the sample customer's credit business has occurred and remains overdue or in arrears;

[0013] When the transfer probability is greater than a first preset threshold, the characteristic indicator of the sample customer is identified as the positive sample;

[0014] When it is detected that the level of the credit asset classification is maintained at the higher level during the preset observation period, the characteristic indicator of the sample customer is identified as a negative sample.

[0015] Optionally, the process of pre-training the prediction model based on the training set as input and the default probability of samples in the training set as training targets includes:

[0016] Collect the acquired positive samples and negative samples to construct a sample set;

[0017] Sampling negative samples from the sample set, and splitting the sample set after sampling into a training set and a validation set;

[0018] The training set is used as input, the default probability of the samples in the training set is used as the training target, and the prediction model is trained using a preset logistic regression algorithm;

[0019] The prediction model obtained through training is verified using the verification set.

[0020] Optionally, the type of the characteristic index includes a continuous independent variable and a discrete independent variable;

[0021] Before taking the training set as input, taking the default probability of the samples in the training set as the training target, and training the prediction model using the preset logistic regression algorithm, the method further includes:

[0022] Eliminate feature indicators in the training set whose missing rate is greater than a second preset threshold;

[0023] Filling missing values ​​for feature indicators whose missing rate is not greater than the second preset threshold and whose type is a continuous independent variable;

[0024] Setting a separate bin for missing values ​​of feature indicators whose missing rate is not greater than the second preset threshold and whose type is a discrete independent variable;

[0025] The characteristic indicators of the continuous independent variables are clustered to obtain a plurality of different categorical variables; wherein each of the categorical variables is set with a separate bin.

[0026] Optionally, also include:

[0027] Calculate the KS value and IV value of each categorical variable;

[0028] The categorical variables whose KS value is less than the third preset threshold and whose IV value is less than the fourth preset threshold are deleted.

[0029] Optionally, also include:

[0030] Taking the categorical variables retained in the training set and the characteristic indicators of the discrete independent variables as target indicators;

[0031] The target indicator is WOE-encoded to obtain the WOE value of the target indicator; wherein the WOE value is used to characterize the default probability of the samples in the training set.

[0032] Optionally, the using the validation set to validate the prediction model obtained through training includes:

[0033] The KS value and AUC value of the training set and the validation set are calculated respectively to verify the model discrimination ability;

[0034] Compare the score values ​​of the training set and the validation set, calculate the PSI value, and verify the inter-period stability;

[0035] The default rate / depth and recall rate / depth variables are calculated using the scoring results of the training set and the validation set to verify the sorting capability.

[0036] A credit risk assessment device, comprising:

[0037] An extraction unit is used to extract a target indicator from the credit behavior information of the customer to be tested in advance; the target indicator is a characteristic indicator that meets a preset condition; the preset condition is that the characteristic indicator is associated with a newly occurring credit behavior;

[0038] A prediction unit is used to input the target indicator of the customer to be tested into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the default probability of samples in the training set as training targets; the samples include pre-acquired positive samples and negative samples; and the output result includes the newly-occurring credit risk of the customer to be tested.

[0039] A computer-readable storage medium includes a stored program, wherein the program executes the credit risk assessment method.

[0040] A credit risk assessment device, comprising: a processor, a memory and a bus; the processor and the memory are connected via the bus;

[0041] The memory is used to store a program, and the processor is used to run the program, wherein the credit risk assessment method is executed when the program is run.

[0042] The technical solution provided by the present application extracts target indicators from the credit behavior information of the customer to be tested in advance. The target indicator is a characteristic indicator that meets the preset conditions, and the preset conditions are: there is a correlation between the characteristic indicator and the newly occurring credit behavior. The target indicator of the customer to be tested is input into the prediction model to obtain the output result of the prediction model. Among them, the prediction model is based on the training set as input, and the default probability of the samples in the training set is used as the training target, and is pre-trained. The samples include positive samples and negative samples obtained in advance. The output results include the newly occurring credit risk of the customer to be tested. It can be seen that by using the solution described in the present application, there is no need to manually consider quantitative and qualitative indicators, and the use of the pre-trained prediction model to evaluate the newly occurring credit risk can effectively improve the accuracy of the newly occurring credit risk evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1a A schematic diagram of a credit risk assessment method provided in an embodiment of the present application;

[0045] Figure 1b A schematic diagram of another credit risk assessment method provided in an embodiment of the present application;

[0046] Figure 2A schematic diagram of another credit risk assessment method provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of the structure of a credit risk assessment device provided in an embodiment of the present application;

[0048] Figure 4 A schematic diagram of the architecture of a credit risk assessment device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] like Figure 1a FIG. 1 is a schematic diagram of a credit risk assessment method provided in an embodiment of the present application, comprising the following steps:

[0051] S101: Extract characteristic indicators from the credit behavior information of sample customers in advance.

[0052] Among them, sample customers include but are not limited to medium-sized enterprises specified by preset standards (such as the enterprise size levels of the four ministries and commissions), and characteristic indicators include but are not limited to: credit asset classification, credit overdue and interest arrears status, basic characteristics of sample customers, industry, region, guarantee status, deposit and loan business processing status, related party status, and changes in financial statements.

[0053] S102: When it is detected that the level of credit asset classification changes from a higher level to a lower level during the preset observation period, the characteristic indicators of the sample customers are marked as positive samples.

[0054] Optionally, when it is detected that the level of the credit asset classification has not changed during the preset observation period, the probability of the sample customer's transition between the dishonest state and the trustworthy state is calculated through the probability transfer matrix; wherein, if the sample customer has dishonest behavior (for example, failure to repay on time leads to arrears of interest, and the duration of the arrears of interest exceeds a preset time), the sample customer is determined to be in a dishonest state, otherwise the sample customer is determined to be in a trustworthy state. The dishonest state can be that the sample customer's credit business occurs and remains in a state of overdue or arrears of interest.

[0055] Specifically, for the sample customers who were in the trustworthy state in the previous period, 99.76% of them were transferred to the trustworthy state in the next period, and 0.23% of them were transferred to the first dishonest state (for example, 1 to 10 days overdue interest). For the sample customers who were in the first dishonest state in the previous period, 30.51% of them were transferred to the trustworthy state in the next period, 5.08% of them were still in the first dishonest state, and 64.41% of them were transferred to the second dishonest state (for example, 10 to 20 days overdue interest). The probability of the second dishonest state changing to the third dishonest state (for example, 20 to 30 days overdue interest), the third dishonest state changing to the fourth dishonest state (for example, 30 to 40 days overdue interest), and the subsequent multi-period transitions were all greater than 90%. The probability transfer matrix showed that when the sample customer was in the second dishonest state, the probability of the sample customer becoming a negative sample tended to be stable.

[0056] When the transfer probability is greater than a first preset threshold, the characteristic indicator of the sample customer is marked as a positive sample;

[0057] It should be noted that the above specific implementation process is only used for illustration.

[0058] The preset observation period is common knowledge familiar to those skilled in the art. Specifically, it can be set to the end of the natural year, and a performance period of more than one year is reserved. For example, if the current year is 2020, the observation period is set to the end of 2018. The level of credit asset classification can usually be regarded as the credit level of the sample customer. The higher the level of credit asset classification, the higher the credit level of the sample customer. The lower the level of credit asset classification, the lower the credit level of the sample customer.

[0059] It should be noted that, given the relatively dispersed distribution of credit periods for medium-sized enterprises, the positive samples selected are all newly occurring during the observation period, rather than using the point-in-time status, which is conducive to identifying the characteristics of newly occurring positive samples.

[0060] S103: When it is detected that the level of credit asset classification remains at a higher level during the preset observation period, the characteristic indicators of the sample customers are marked as negative samples.

[0061] Among them, when it is detected that the level of credit asset classification remains at a higher level during the preset observation period, the positive samples determined by the probability transfer matrix are eliminated and marked as negative samples.

[0062] S104: Collect the obtained positive samples and negative samples to construct a sample set.

[0063] S105: Sampling negative samples from the sample set, and splitting the sample set into a training set and a validation set.

[0064] Among them, the purpose of sampling negative samples from the sample set is to increase the concentration of positive samples in the sample set.

[0065] S106: Eliminate feature indicators in the training set whose missing rate is greater than a second preset threshold, and fill in missing values ​​for feature indicators whose missing rate is not greater than the second preset threshold.

[0066] The specific implementation method of filling in missing values ​​is common knowledge familiar to those skilled in the art, for example, filling in missing values ​​using the median.

[0067] It should be noted that the types of characteristic indicators include continuous independent variables and discrete independent variables. Specifically, the median can be used to fill missing values ​​for characteristic indicators whose missing rate is not greater than the second preset threshold and whose type is a continuous independent variable, and separate bins can be set for missing values ​​of characteristic indicators whose missing rate is not greater than the second preset threshold and whose type is a discrete independent variable.

[0068] S107: Clustering the characteristic indicators of the continuous independent variables in the training set to obtain a plurality of different categorical variables.

[0069] Among them, each categorical variable is set with a separate bin.

[0070] S108: Calculate the KS value and IV value of each categorical variable.

[0071] The so-called KS value and IV value, as well as the calculation methods of the KS value and IV value, are common knowledge familiar to those skilled in the art and will not be elaborated here.

[0072] S109: Delete the categorical variables whose KS values ​​are less than the third preset threshold and whose IV values ​​are less than the fourth preset threshold.

[0073] S110: Perform WOE encoding on the target indicator to obtain the WOE value of the target indicator.

[0074] Among them, the target indicator includes the categorical variables retained in the training set and the characteristic indicators of the discrete independent variable type. In addition, the WOE value includes the proportion of positive samples in the bin to which the target indicator belongs to all samples in the training set, and the proportion of negative samples in all samples in the training set. In the embodiment of the present application, the WOE value is used to characterize the default probability of the samples in the training set. The samples with a default probability greater than the fifth preset threshold value indicate that the characteristic indicator indicated by the sample is associated with the newly occurring credit behavior.

[0075] It should be noted that the specific implementation method and implementation principle of WOE coding are common knowledge familiar to those skilled in the art and will not be described in detail here.

[0076] S111: Take the training set as input, take the default probability of the samples in the training set as the training target, and use the preset logistic regression algorithm to train the prediction model.

[0077] Among them, the independent variables are screened by the logistic regression algorithm, which is a retail business risk measurement method commonly used in the industry. Practice has proved that this method has wide applicability and rationality. The logistic regression algorithm has its unique advantages in processing data of two-category dependent variables, and the assumptions about data distribution are weak. At the same time, the algorithm has good stability, high relative accuracy, and is easy to explain and develop.

[0078] In the embodiment of the present application, the basic assumption of the logistic regression algorithm is that the logarithm of the ratio of positive samples (good customers) to negative samples (bad customers) is the linear regression of their attributes:

[0079]

[0080] The parameters shown in formula (1) are common knowledge familiar to those skilled in the art and will not be described in detail here.

[0081] Specifically, the Backward method can be used to perform backward regression on the prediction model.

[0082] S112: Use the validation set to validate the trained prediction model.

[0083] Specifically, the prediction model obtained through training is verified using the validation set, including but not limited to:

[0084] 1. Calculate the KS value and AUC value of the training set and the validation set respectively to verify the model's ability to distinguish;

[0085] 2. Compare the scores of the training set and the validation set (for example, the training set is the score at the end of 2019, and the validation set is the score at the end of 2020. Check whether the proportions of each segment of the two-period score are stable) to calculate the PSI value and verify the inter-period stability;

[0086] 3. Using the scoring results of the training set and the validation set, calculate the default rate / depth and recall rate / depth variables to verify the sorting ability.

[0087] It should be noted that the above specific implementation process is only used for illustration. In addition, the KS value, AUC value, PSI value, default rate / depth, and recall rate / depth variables are all common knowledge familiar to those skilled in the art and will not be described in detail here.

[0088] Specifically, the process shown in S101 to S112 above can be found in Figure 1b shown.

[0089] S113: Extract target indicators from the credit behavior information of the customer to be tested in advance.

[0090] S114: Input the target indicator of the customer to be tested into the prediction model to obtain the output result of the prediction model.

[0091] Among them, the output results include the default probability of the customer to be tested (i.e., the newly occurring credit risk).

[0092] In summary, based on the scheme shown in this embodiment, the problem of fewer positive samples for medium-sized enterprises can be solved, so that an independent prediction model can be established for medium-sized enterprises, which is conducive to the identification and prediction of default characteristics of medium-sized enterprises. In addition, the selection of positive samples is all new during the performance period, which is different from the traditional definition of positive samples, and improves the model's prediction ability for new credit default customers. It can be seen that using the scheme described in this embodiment, there is no need to manually consider quantitative and qualitative indicators, and the use of the pre-trained prediction model to conduct credit risk assessment of newly occurring credit can effectively improve the accuracy of credit risk assessment of newly occurring credit.

[0093] It should be noted that S101 mentioned in the above embodiment is an optional specific implementation of the credit risk assessment method described in this application. In addition, S102 mentioned in the above embodiment is also an optional specific implementation of the credit risk assessment method described in this application. To this end, the process mentioned in the above embodiment can be summarized as follows: Figure 2 The method shown.

[0094] like Figure 2 FIG. 1 is a schematic diagram of another credit risk assessment method provided in an embodiment of the present application, comprising the following steps:

[0095] S201: Extract target indicators from the credit behavior information of the customer to be tested in advance.

[0096] Among them, the target indicator is a characteristic indicator that meets the preset conditions; the preset conditions are: there is a correlation between the characteristic indicator and the newly occurring credit behavior.

[0097] S202: Input the target indicator of the customer to be tested into the prediction model to obtain the output result of the prediction model.

[0098] The prediction model is pre-trained based on the training set as input and the default probability of the samples in the training set as the training target. The samples include pre-acquired positive samples and negative samples. The output results include the newly occurred credit risk of the customer to be tested.

[0099] In summary, based on the scheme shown in this embodiment, the problem of fewer positive samples for medium-sized enterprises can be solved, so that an independent prediction model can be established for medium-sized enterprises, which is conducive to the identification and prediction of default characteristics of medium-sized enterprises. In addition, the selection of positive samples is all new during the performance period, which is different from the traditional definition of positive samples, and improves the model's prediction ability for new credit default customers. It can be seen that using the scheme described in this embodiment, there is no need to manually consider quantitative and qualitative indicators, and the use of the pre-trained prediction model to conduct credit risk assessment of newly occurring credit can effectively improve the accuracy of credit risk assessment of newly occurring credit.

[0100] Corresponding to the credit risk assessment method provided in the above-mentioned embodiment of the present application, the embodiment of the present application also provides a credit risk assessment device.

[0101] like Figure 3 FIG. 1 is a schematic diagram of the structure of a credit risk assessment device provided in an embodiment of the present application, including:

[0102] The extraction unit 100 is used to extract the target indicator from the credit behavior information of the customer to be tested in advance; the target indicator is a characteristic indicator that meets the preset condition; the preset condition is: there is a correlation between the characteristic indicator and the newly occurring credit behavior.

[0103] The prediction unit 200 is used to input the target indicators of the customer to be tested into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the probability of default as a training target; the training set includes pre-acquired positive samples and negative samples; the output result includes the newly-occurring credit risk of the customer to be tested.

[0104] The prediction unit 200 is used to obtain the process of positive samples and negative samples, including: extracting characteristic indicators from the credit behavior information of the sample customer in advance; wherein the characteristic indicators include credit asset classification, credit overdue, and interest arrears; in the case where it is detected that the level of the credit asset classification changes from a higher level to a lower level during the preset observation period, the characteristic indicators of the sample customer are identified as positive samples; in the case where it is detected that the level of the credit asset classification does not change during the preset observation period, the transition probability of the sample customer between the untrustworthy state and the trustworthy state is calculated through a probability transition matrix; wherein if the sample customer has untrustworthy behavior, it is determined that the sample customer is in the untrustworthy state, otherwise it is determined that the sample customer is in the trustworthy state, and the untrustworthy state may be that the sample customer's credit business occurs and remains overdue or in arrears; in the case where the transition probability is greater than a first preset threshold, the characteristic indicators of the sample customer are identified as positive samples; in the case where it is detected that the level of the credit asset classification remains at the higher level during the preset observation period, the characteristic indicators of the sample customer are identified as negative samples.

[0105] The prediction unit 200 is used to pre-train a prediction model based on a training set as input and the default probability of samples in the training set as a training target, including: collecting the acquired positive samples and negative samples to construct a sample set; sampling negative samples from the sample set, and splitting the sampled sample set into a training set and a validation set; using the training set as input and the default probability of samples in the training set as a training target, and training the prediction model using a preset logistic regression algorithm; and validating the trained prediction model using the validation set.

[0106] The types of feature indicators include continuous independent variables and discrete independent variables. The prediction unit 200 is also used to: eliminate feature indicators in the training set whose missing rate is greater than a second preset threshold; fill in missing values ​​for feature indicators whose missing rate is not greater than the second preset threshold and whose type is a continuous independent variable; set separate bins for missing values ​​of feature indicators whose missing rate is not greater than the second preset threshold and whose type is a discrete independent variable; cluster feature indicators that are continuous independent variables to obtain multiple different classification variables; wherein a separate bin is set for each classification variable.

[0107] The prediction unit 200 is further used to: calculate the KS value and IV value of each categorical variable; and delete the categorical variables whose KS value is less than the third preset threshold and whose IV value is less than the fourth preset threshold.

[0108] The prediction unit 200 is also used to: use the categorical variables retained in the training set and the characteristic indicators of the discrete independent variables as target indicators; perform WOE encoding on the target indicators to obtain the WOE value of the target indicators; wherein the WOE value is used to characterize the default probability of the samples in the training set.

[0109] The prediction unit 200 is used to verify the prediction model obtained through training using the validation set, including: respectively calculating the KS value and AUC value of the training set and the validation set to verify the model's ability to distinguish; comparing the scoring values ​​of the training set and the validation set to calculate the PSI value to verify the inter-period stability; using the scoring results of the training set and the validation set to calculate the default rate / depth and recall rate / depth variables to verify the sorting ability.

[0110] In summary, based on the scheme shown in this embodiment, the problem of fewer positive samples for medium-sized enterprises can be solved, so that an independent prediction model can be established for medium-sized enterprises, which is conducive to the identification and prediction of default characteristics of medium-sized enterprises. In addition, the selection of positive samples is all new during the performance period, which is different from the traditional definition of positive samples, and improves the model's prediction ability for new credit default customers. It can be seen that using the scheme described in this embodiment, there is no need to manually consider quantitative and qualitative indicators, and the use of the pre-trained prediction model to conduct credit risk assessment of newly occurring credit can effectively improve the accuracy of credit risk assessment of newly occurring credit.

[0111] The present application also provides a computer-readable storage medium, which includes a stored program, wherein the program executes the credit risk assessment method provided by the present application.

[0112] The present application also provides a credit risk assessment device, such as Figure 4 As shown, it includes: a processor 401, a memory 402 and a bus 403. The processor 401 is connected to the memory 402 via the bus 403, the memory 402 is used to store programs, and the processor 401 is used to run the program, wherein when the program is run, the credit risk assessment method provided by the present application is executed, including the following steps:

[0113] Extracting target indicators from the credit behavior information of the customer to be tested in advance; the target indicators are characteristic indicators that meet preset conditions; the preset conditions are: the characteristic indicators are associated with the newly occurring credit behavior;

[0114] The target indicators of the customer to be tested are input into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the default probability of samples in the training set as training targets; the samples include pre-acquired positive samples and negative samples; the output result includes the newly-occurred credit risk of the customer to be tested.

[0115] Optionally, the process of obtaining the positive sample and the negative sample includes:

[0116] Extract characteristic indicators from the credit behavior information of sample customers in advance; wherein the characteristic indicators include credit asset classification, credit overdue, and interest arrears status;

[0117] If it is detected that the credit asset classification level changes from a higher level to a lower level during the preset observation period, the characteristic indicators of the sample customers will be marked as positive samples;

[0118] When it is detected that the level of the credit asset classification has not changed during the preset observation period, the probability of the sample customer's transition between the dishonest state and the trustworthy state is calculated through the probability transition matrix; wherein, if the sample customer has dishonest behavior, the sample customer is determined to be in the dishonest state, otherwise, the sample customer is determined to be in the trustworthy state, and the dishonest state may be that the sample customer's credit business has occurred and remains overdue or in arrears;

[0119] When the transfer probability is greater than a first preset threshold, the characteristic indicator of the sample customer is marked as a positive sample;

[0120] When it is detected that the level of the credit asset classification is maintained at the higher level during the preset observation period, the characteristic indicator of the sample customer is identified as the negative sample.

[0121] Optionally, the process of pre-training the prediction model based on the training set as input and the default probability of samples in the training set as training targets includes:

[0122] Collect the acquired positive samples and negative samples to construct a sample set;

[0123] Sampling negative samples from the sample set, and splitting the sample set after sampling into a training set and a validation set;

[0124] The training set is used as input, the default probability of the samples in the training set is used as the training target, and the prediction model is trained using a preset logistic regression algorithm;

[0125] The prediction model obtained through training is verified using the verification set.

[0126] Optionally, the type of the characteristic index includes a continuous independent variable and a discrete independent variable;

[0127] Before taking the training set as input, taking the default probability of samples in the training set as training targets, and using a preset logistic regression algorithm to train the prediction model, the method further includes:

[0128] Eliminate feature indicators in the training set whose missing rate is greater than a second preset threshold;

[0129] Filling missing values ​​for feature indicators whose missing rate is not greater than the second preset threshold and whose type is a continuous independent variable;

[0130] Setting a separate bin for missing values ​​of feature indicators whose missing rate is not greater than the second preset threshold and whose type is a discrete independent variable;

[0131] The characteristic indicators of the continuous independent variables are clustered to obtain a plurality of different categorical variables; wherein each of the categorical variables is set with a separate bin.

[0132] Optionally, also include:

[0133] Calculate the KS value and IV value of each categorical variable;

[0134] The categorical variables whose KS value is less than the preset third threshold and whose IV value is less than the preset fourth threshold are deleted.

[0135] Optionally, also include:

[0136] Taking the categorical variables retained in the training set and the characteristic indicators of the discrete independent variables as target indicators;

[0137] The target indicator is WOE-encoded to obtain the WOE value of the target indicator; wherein the WOE value is used to characterize the default probability of the samples in the training set.

[0138] Optionally, the using the validation set to validate the prediction model obtained through training includes:

[0139] The KS value and AUC value of the training set and the validation set are calculated respectively to verify the model discrimination ability;

[0140] Compare the score values ​​of the training set and the validation set, calculate the PSI value, and verify the inter-period stability;

[0141] The default rate / depth and recall rate / depth variables are calculated using the scoring results of the training set and the validation set to verify the sorting capability.

[0142] If the functions described in the method of the embodiment of the present application are implemented in the form of software functional units and sold or used as independent products, they can be stored in a storage medium readable by a computing device. Based on this understanding, the part of the embodiment of the present application that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions to enable a computing device (which can be a personal computer, server, mobile computing device or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0143] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0144] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A credit risk assessment method, characterized in that: include: Extract target indicators from the credit behavior information of the customers to be tested in advance; The target indicator is a characteristic indicator that meets the preset conditions, and the characteristic indicator includes credit asset classification, credit overdue, and interest arrears status; the preset condition is that the characteristic indicator is associated with the newly occurred credit behavior; Input the target index of the customer to be tested into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the default probability of samples in the training set as training targets; the samples include pre-acquired positive samples and negative samples; the output result includes the newly-occurring credit risk of the customer to be tested, and the newly-occurring credit risk is expressed based on the default probability; The process of obtaining the positive sample and the negative sample includes: Extract characteristic indicators from the credit behavior information of sample customers in advance; When it is detected that the level of the credit asset classification changes from a higher level to a lower level during the preset observation period, the characteristic indicator of the sample customer is identified as a positive sample; When it is detected that the level of the credit asset classification has not changed during the preset observation period, the probability of the sample customer's transition between the dishonest state and the trustworthy state is calculated through the probability transition matrix; wherein, if the sample customer has dishonest behavior, the sample customer is determined to be in the dishonest state; otherwise, the sample customer is determined to be in the trustworthy state, and the dishonest state means that the sample customer's credit business has occurred and remains overdue or in arrears; When the transfer probability is greater than a first preset threshold, identifying the characteristic indicator of the sample customer as the positive sample; When it is detected that the level of the credit asset classification is maintained at the higher level during the preset observation period, the characteristic indicator of the sample customer is marked as a negative sample; The process of pre-training the prediction model based on the training set as input and the default probability of the samples in the training set as the training target includes: Collect the acquired positive samples and negative samples to construct a sample set; Sampling negative samples from the sample set, and splitting the sample set after sampling into a training set and a validation set; The training set is used as input, the default probability of the samples in the training set is used as the training target, and the prediction model is trained using a preset logistic regression algorithm; The prediction model obtained through training is verified using the verification set.

2. The method according to claim 1, characterized in that The types of characteristic indicators include continuous independent variables and discrete independent variables; Before taking the training set as input, taking the default probability of the samples in the training set as the training target, and training the prediction model using the preset logistic regression algorithm, the method further includes: Eliminate feature indicators in the training set whose missing rate is greater than a second preset threshold; Filling missing values ​​for feature indicators whose missing rate is not greater than the second preset threshold and whose type is a continuous independent variable; Setting a separate bin for missing values ​​of feature indicators whose missing rate is not greater than the second preset threshold and whose type is a discrete independent variable; The characteristic indicators of the continuous independent variables are clustered to obtain a plurality of different categorical variables; wherein each of the categorical variables is set with a separate bin.

3. The method according to claim 2, characterized in that Also includes: Calculate the KS value and IV value of each categorical variable; The categorical variables whose KS value is less than the third preset threshold and whose IV value is less than the fourth preset threshold are deleted.

4. The method according to claim 2, characterized in that: Also includes: Taking the categorical variables retained in the training set and the characteristic indicators of the discrete independent variables as target indicators; The target indicator is WOE-encoded to obtain the WOE value of the target indicator; wherein the WOE value is used to characterize the default probability of the samples in the training set.

5. The method according to claim 1, characterized in that The using the verification set to verify the prediction model obtained through training includes: The KS value and AUC value of the training set and the validation set are calculated respectively to verify the model discrimination ability; Compare the score values ​​of the training set and the validation set, calculate the PSI value, and verify the inter-period stability; The default rate / depth and recall rate / depth variables are calculated using the scoring results of the training set and the validation set to verify the sorting capability.

6. A credit risk assessment device, characterized in that: include: An extraction unit, used for extracting target indicators from the credit behavior information of the customer to be tested in advance; The target indicator is a characteristic indicator that meets the preset conditions, and the characteristic indicator includes credit asset classification, credit overdue, and interest arrears status; the preset condition is that the characteristic indicator is associated with the newly occurred credit behavior; A prediction unit, used for inputting the target index of the customer to be tested into the prediction model to obtain the output result of the prediction model; wherein the prediction model is pre-trained based on a training set as input and the default probability of samples in the training set as training targets; the samples include pre-acquired positive samples and negative samples; the output result includes the newly-occurring credit risk of the customer to be tested, and the newly-occurring credit risk is expressed based on the default probability; The prediction unit is also used to extract characteristic indicators from the credit behavior information of the sample customer in advance; when it is detected that the level of the credit asset classification changes from a higher level to a lower level during the preset observation period, the characteristic indicator of the sample customer is identified as a positive sample; when it is detected that the level of the credit asset classification does not change during the preset observation period, the transition probability of the sample customer between the untrustworthy state and the trustworthy state is calculated through a probability transition matrix; wherein, if the sample customer has untrustworthy behavior, it is determined that the sample customer is in the untrustworthy state, otherwise it is determined that the sample customer is in the trustworthy state, and the untrustworthy state is that the sample customer's credit business occurs and remains overdue or in arrears; when the transition probability is greater than a first preset threshold, the characteristic indicator of the sample customer is identified as a positive sample; when it is detected that the level of the credit asset classification remains at the higher level during the preset observation period, the characteristic indicator of the sample customer is identified as the negative sample; The prediction unit is further used to collect the positive samples and the negative samples obtained to construct a sample set; sample negative samples from the sample set, and split the sample set after sampling into a training set and a validation set; use the training set as input, take the default probability of the samples in the training set as the training target, and use a preset logistic regression algorithm to train the prediction model; and use the validation set to verify the prediction model obtained through training.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein the program executes the credit risk assessment method according to any one of claims 1 to 5.

8. A credit risk assessment device, characterized in that: include: processor, memory, and bus; The processor is connected to the memory via the bus; The memory is used to store programs, and the processor is used to run programs, wherein the program executes the credit risk assessment method according to any one of claims 1 to 5 when running.

Images