Binary tree-based credit risk dynamic attribution analysis method

Abstract

The invention provides a dynamic credit risk attribution analysis method based on a binary tree, and belongs to the technical field of financial data, and the method specifically comprises the steps: determining the composition data of historical credit users corresponding to a surprise degree related group in an index evaluation time period, determining an interpretation degree threshold value of the surprise degree associated groups by combining the number of the surprise degree associated groups and the change data of the credit risk indexes in the index evaluation period, and constructing a binary tree on the basis of the surprise degree associated groups and the interpretation degree threshold value; and binary tree leaf sub-nodes are generated on the basis of the interpretation degrees and surprise degrees of different surprise degree associated groups, and the attribution result of the credit risk index is generated, so that the accuracy of attribution analysis is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of financial data, and in particular relates to a credit risk dynamic attribution analysis method based on a binary tree. Background Art

[0002] In the field of credit risk control, the active payment rate, overdue rate and recovery rate are key indicators to measure the company's operating level and risk control level. However, when these indicators fluctuate abnormally, the company needs to quickly identify which customer group caused this result. Traditional attribution analysis mainly relies on manual monitoring, analysis and decision-making, which has problems such as low efficiency and untimely risk response.

[0003] In order to automatically realize the attribution analysis of abnormal credit indicators, the invention patent application CN202411974941.5 "A method and system for attribution analysis of overdue indicator changes" determines the characteristics of related indicators based on the changes in credit indicators, and performs attribution analysis of credit indicator changes based on the sorting results of the correlation coefficients of the related indicator characteristics, which improves the accuracy of the attribution analysis processing. However, there are the following technical problems: When conducting attribution analysis, since there are a large number of credit indicators and different credit indicators have certain correlations, if the drill-down dimension method is used to implement the analysis and processing of related credit indicators, it will inevitably lead to the processing efficiency and difficulty of the attribution analysis being unable to meet the requirements.

[0004] In order to solve the above technical problems, the present application provides a credit risk dynamic attribution analysis method based on a binary tree. Summary of the invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions: Specifically, the present application provides a credit risk dynamic attribution analysis method based on a binary tree, which specifically includes: S1 divides historical credit users into multiple customer groups by using credit product types, and determines the indicator stability period of the credit risk indicator based on the changes in the credit risk indicator in different unit time periods; S2: obtaining changes in the correlation indicators of the customer groups in different indicator stability periods, and determining the surprise-related groups in the customer groups based on the changes in the correlation indicators; S3: determining the composition data of the historical credit users corresponding to the surprise degree associated group in the index evaluation period, and determining the explanation threshold of the surprise degree associated group in combination with the number of the surprise degree associated group and the change data of the credit risk index in the index evaluation period; S4 constructs a binary tree based on the surprise degree associated groups and the explanation degree threshold, and generates binary tree leaf nodes and generates attribution results of credit risk indicators based on the explanation degrees and surprise degrees of different surprise degree associated groups.

[0006] The beneficial effects of the present invention are: The surprise-related group in the customer group is determined by the changes in the related indicators during the stable indicator period, which enables the screening of customer groups with frequent changes in related indicators. This also lays the foundation for excluding customer groups with frequent changes in related indicators when conducting attribution analysis, thereby ensuring the processing efficiency of attribution analysis.

[0007] The generation of binary tree leaf nodes and the attribution results of credit risk indicators are processed based on the explanatory power and surprise degree of different surprise association groups. Full consideration is given to the differences in the degree of correlation between the changes in association indicators and the changes in credit risk indicators of different precision association groups when using the binary tree for attribution analysis. Differentiated binary tree leaf nodes are generated, which improves the efficiency and accuracy of the generation and processing of binary tree leaf nodes, and also improves the efficiency of attribution analysis processing.

[0008] A further technical solution is that the type of credit product is determined based on credit products developed for different groups by a credit service platform.

[0009] A further technical solution is to divide historical credit users into multiple customer groups, including: Customers belonging to the same type of credit product are grouped into the same customer group.

[0010] A further technical solution is that the credit risk indicators include the utilization rate, the overdue rate and the recovery rate.

[0011] A further technical solution is that the method for determining the stable period of the credit risk indicator is: Determine the benchmark value in different unit periods by taking the average value of the credit risk indicator on different dates in different unit periods; Determine the index change date in the date according to the deviation between the credit risk index of different dates in the unit period and the benchmark value; Whether the unit period is a stable period of the credit risk indicator is determined based on the proportion of the number of dates on which the indicator changes.

[0012] A further technical solution is that the indicator change date is a date when the deviation of the credit indicator from the benchmark value is not within a preset indicator deviation range.

[0013] A further technical solution is that, when the proportion of the number of indicator change dates in the unit time period is greater than the proportion of the number of preset change dates, it is determined that the unit time period does not belong to the indicator stability period of the credit risk indicator.

[0014] A further technical solution is to construct a binary tree, which specifically includes: Obtain the corresponding surprise-related index data, explanation threshold, and customer group proportion threshold; If the proportion of historical credit users in the surprise-related group is less than the customer group proportion threshold, the explanation and surprise of the customer group are calculated, and the customer group is used as a binary tree leaf node; If the proportion of historical credit users in the surprise-related group is not less than the customer group proportion threshold, Calculate the explanatory power and surprise power of the associated index of the customer group, and when the explanatory power of the surprise power associated group does not meet the constraint condition of the explanatory power threshold, use the surprise power associated group as a binary tree leaf node; When the explanatory power of the surprise degree associated group meets the constraint condition of the explanatory power threshold, the surprise degree associated group is secondary divided into subgroups based on the similarity of basic information of historical credit users of subgroups of the surprise degree associated group, and binary tree leaf nodes are constructed based on the subgroups whose surprise degrees meet the surprise degree threshold, and a decision tree is constructed using the binary tree leaf nodes.

[0015] A further technical solution is to classify historical credit users whose basic information deviations in a certain dimension are within a preset deviation range into the same sub-group.

[0016] A further technical solution is that the basic information includes occupation, income, region, credit limit, expenditure data and client.

[0017] A further technical solution is that the explanatory power of the association index of the surprise association group is determined based on the ratio of the deviation between the average value of the association index of the surprise association group in the evaluation target period and the association index in the stability index period, and the deviation between the average value of the association index of the credit risk indicator in the evaluation target period and the stability index period.

[0018] A further technical solution is that the method for determining the surprise degree of the surprise degree associated group is: Determine the benchmark values ​​in different unit periods based on the average values ​​of the credit risk indicators on different dates in different unit periods, and determine the indicator change dates on the dates based on the deviations between the credit risk indicators on different dates in the unit period and the benchmark values; The surprise degree of the customer group is determined according to the ratio of the number of indicator change dates in the evaluation target period to the number of indicator change dates in different indicator stable periods.

[0019] A further technical solution is to generate and process the attribution results of the credit risk indicators, specifically including: The customer group whose explanatory degree of the binary tree leaf node meets the explanatory degree threshold is used as the output result of the credit risk indicator.

[0020] Other features and advantages will be described in the following description. The objects and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0023] Figure 1 It is a flowchart of a credit risk dynamic attribution analysis method based on a binary tree; Figure 2 It is a flow chart of a method for determining an indicator stability period of a credit risk indicator; Figure 3 is a flow chart of a method for determining a surprise-related group in a customer group; Figure 4 It is a flowchart of a method for determining an explanation threshold of a surprise degree association group. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0025] In the daily work of credit institutions, once there are changes in credit risk indicators such as the application rate, overdue rate and recovery rate, due to the large number of customer groups, credit institutions are often unable to accurately determine which customer groups' related indicators have changed, resulting in the change in credit risk indicators, and thus are unable to generate targeted intervention measures.

[0026] Specifically, if there is an abnormality in the Apple client, it will inevitably lead to a lower utilization rate of loans for credit users whose clients are Apple, which may cause changes in the credit institution's withdrawal rate. Therefore, if the abnormal situation of the above-mentioned customer groups cannot be discovered in time and the client cannot be repaired in time, it will be impossible to ensure that the withdrawal rate of the credit structure remains within a reasonable range.

[0027] In this application, surprise degree and explanation degree are used as variables, and a binary tree model is used to screen customer groups with large changes in related indicators, so that abnormal situations can be identified more quickly.

[0028] The benchmark values ​​in different unit time periods are determined by taking the average values ​​of the credit risk indicators on different dates in different unit time periods; the indicator change dates in the dates are determined according to the deviations of the credit risk indicators on different dates in the unit time period from the benchmark values; the surprise degree of the customer group is determined according to the ratio of the proportion of the number of indicator change dates in the evaluation target time period to the proportion of the number of indicator change dates in different indicator stable time periods.

[0029] The degree of surprise reflects the degree of consistency between the daily changes in the associated indicators and the changes in the indicator evaluation period.

[0030] The explanatory power of the correlation indicator is determined based on the ratio of the deviation between the average value of the correlation indicator of the surprise correlation group in the evaluation target period and the average value of the correlation indicator in the stable index period, and the deviation between the average value of the correlation indicator of the credit risk indicator in the evaluation target period and the stable index period, which reflects the changing correlation between the correlation indicator and the credit risk indicator.

[0031] Specifically, Figure 1 As shown, the present application provides a credit risk dynamic attribution analysis method based on a binary tree, which specifically includes: S1 divides historical credit users into multiple customer groups by using credit product types, and determines the indicator stability period of the credit risk indicator based on the changes in the credit risk indicator in different unit time periods; The benchmark values ​​in different unit periods are determined by taking the average values ​​of the credit risk indicators on different dates in different unit periods. The indicator change date is the date when the deviation between the credit indicator and the benchmark value is not within the preset indicator deviation range. When the number of indicator change dates in a unit period accounts for more than 0.3, it is determined that the unit period does not belong to the indicator stability period of the credit risk indicator. S2: obtaining changes in the correlation indicators of the customer groups in different indicator stability periods, and determining the surprise-related groups in the customer groups based on the changes in the correlation indicators; The reference indicator data in the indicator stable period is determined according to the average values ​​of the associated indicators on different dates in the indicator stable period. When there is an indicator stable period in which the deviation of the reference indicator data between the indicator stable periods and other indicator stable periods does not meet the requirements, it is determined that the customer group does not belong to the surprise degree associated group.

[0032] S3: determining the composition data of the historical credit users corresponding to the surprise degree associated group in the index evaluation period, and determining the explanation threshold of the surprise degree associated group in combination with the number of the surprise degree associated group and the change data of the credit risk index in the index evaluation period; Determine the association weight coefficients of different surprise degree association groups according to the proportion of the number of historical credit users in different surprise degree association groups, and use the surprise degree association groups whose association weight coefficients are greater than the preset weight coefficients as the screening association groups; The deviation of the credit risk indicator in the indicator evaluation period from a preset benchmark risk indicator threshold is determined based on the change data of the credit risk indicator in the indicator evaluation period; the deviation rate is determined based on the ratio of the deviation amount to the preset benchmark risk indicator threshold; the data matching coefficient is determined based on the ratio of the number of the screened associated groups to the number of the customer groups; the compensation factor is determined based on the ratio of the deviation rate to the data matching coefficient; the compensation threshold is determined based on the product of the preset explanatory threshold and the compensation factor; and the explanatory threshold of the surprise associated group is determined based on the difference between the compensation threshold and the preset explanatory threshold.

[0033] S4 constructs a binary tree based on the surprise degree associated groups and the explanation degree threshold, and generates binary tree leaf nodes and generates attribution results of credit risk indicators based on the explanation degrees and surprise degrees of different surprise degree associated groups.

[0034] Furthermore, the credit product type is determined based on the credit products developed by the credit service platform for different groups.

[0035] Specifically, historical credit users are divided into multiple customer groups, including: Customers belonging to the same type of credit product are grouped into the same customer group.

[0036] It should be noted that the credit risk indicators include the liquidity ratio, overdue rate and recovery rate.

[0037] Specifically, Figure 2 As shown, the method for determining the stable period of the credit risk indicator is: Determine the benchmark value in different unit periods by taking the average value of the credit risk indicator on different dates in different unit periods; Determine the index change date in the date according to the deviation between the credit risk index of different dates in the unit period and the benchmark value; Whether the unit period is a stable period of the credit risk indicator is determined based on the proportion of the number of dates on which the indicator changes.

[0038] Furthermore, the indicator change date is a date when the deviation of the credit indicator from the benchmark value is not within a preset indicator deviation range.

[0039] It can be understood that when the proportion of the number of indicator change dates in the unit time period is greater than the proportion of the number of preset change dates, it is determined that the unit time period does not belong to the indicator stability period of the credit risk indicator.

[0040] Specifically, Figure 3 As shown, the method for determining the surprise degree associated group in the customer group is: Determine the associated indicators on different dates in the indicator stable period based on the changes in the associated indicators of the customer groups in the indicator stable period; Determining a change date in a stable period of the indicator based on a change amount of an indicator associated with adjacent dates; Whether the customer group is a surprise-related group is determined by the proportion of the number of change dates in different indicator stable periods.

[0041] Furthermore, the change date is a date on which the change amount of the associated indicator with the adjacent date is greater than a preset indicator change amount threshold.

[0042] It should be noted that determining whether the customer group is a surprise-related group based on the proportion of the number of change dates in different indicator stable periods specifically includes: The proportion of the number of change dates in different indicator stable periods is used as the coefficient of change of the associated indicator in different indicator stable periods; The average value of the coefficient of variation of the associated indicators in different indicator stability periods is taken as the average coefficient of variation; When the average variation coefficient is greater than a preset variation coefficient threshold, it is determined that the customer group does not belong to a surprise degree associated group.

[0043] Optionally, the method for determining the surprise degree associated group in the customer group is: Determine the associated indicators on different dates in the indicator stable period based on the changes in the associated indicators of the customer groups in the indicator stable period; Determine the reference indicator data in the indicator stable period according to the average values ​​of the associated indicators on different dates in the indicator stable period; Whether the customer group is a surprise-related group is determined based on the deviation of the reference indicator data between different indicator stable periods.

[0044] Further, determining whether the customer group is a surprise-related group based on the deviation of the reference indicator data between different indicator stable periods specifically includes: When there is an indicator stable period in which the deviation amount of the reference indicator data between other indicator stable periods does not meet the requirements, it is determined that the customer group does not belong to the surprise degree associated group.

[0045] Optionally, the method for determining the surprise degree associated group in the customer group is: S21: determining the associated indicators on different dates in the stable indicator period according to the change of the associated indicators of the customer group in the stable indicator period, determining the reference indicator data in the stable indicator period according to the average value of the associated indicators on different dates in the stable indicator period, and determining the period variation coefficient of the associated indicators of the customer group according to the deviation of the reference indicator data between different stable indicator periods; S22: determining the change date in the indicator stability period based on the change amount of the associated indicator of the adjacent date, and determining the associated indicator change coefficient of different indicator stability periods according to the proportion of the number of change dates in different indicator stability periods; S23 determines the stability deviation coefficient of the associated indicator of the customer group based on the associated indicator variation coefficient of different indicator stability periods and the period variation coefficient, and uses the stability deviation coefficient to determine whether the customer group is a surprise associated group.

[0046] Further, when the stability deviation coefficient does not meet the requirement, it is determined that the customer group does not belong to the surprise degree associated group.

[0047] Optionally, the above step S21 includes the following contents: S211 determines the associated indicators on different dates in the stable indicator period according to the changes in the associated indicators of the customer group in the stable indicator period, and determines the reference indicator data in the stable indicator period according to the average values ​​of the associated indicators on different dates in the stable indicator period. When there is a stable indicator period whose deviation amount of the reference indicator data between other stable indicator periods does not meet the requirements, it is determined that the customer group does not belong to the surprise degree associated group. When there is no stable indicator period whose deviation amount of the reference indicator data between other stable indicator periods does not meet the requirements, it proceeds to step S212. S212 determines the data deviation of the reference indicator data of different indicator stable periods based on the deviation between the reference indicator data of the indicator stable period and the reference indicator data of other indicator stable periods. When there is an indicator stable period where the data deviation does not meet the requirement, the process proceeds to step S213. When there is no indicator stable period where the data deviation does not meet the requirement, the process proceeds to step S214. S213: When the number of stable time periods of the indicator in which the data deviation amount does not meet the requirement is greater than the number of preset stable time periods, it is determined that the customer group does not belong to the surprise degree associated group; when the number of stable time periods of the indicator in which the data deviation amount does not meet the requirement is not greater than the number of preset stable time periods, the process proceeds to step S214; S214 determines the time period variation coefficient of the associated indicator of the customer group based on the deviation of the reference indicator data between different indicator stability periods. When the time period variation coefficient of the associated indicator of the customer group does not meet the requirements, it is determined that the customer group does not belong to the surprise degree associated group. When the time period variation coefficient of the associated indicator of the customer group meets the requirements, the process proceeds to step S22.

[0048] Optionally, the above step S22 includes the following contents: S221: Based on the change amount of the associated index with the adjacent dates, when it is determined that there is no change date in the different index stable periods, it is determined that the customer group belongs to the surprise degree associated group. When there is a change date in the different index stable periods, the process proceeds to step S222; S222: Determine the correlation index variation coefficients of different indicator stable periods according to the proportion of the number of change dates in different indicator stable periods; when the average value of the correlation index variation coefficients of different indicator stable periods does not meet the requirement, determine that the customer group belongs to the surprise correlation group; when the average value of the correlation index variation coefficients of different indicator stable periods meets the requirement, proceed to step S223; S223: When there is a stable index period in which the coefficient of change of the associated index does not meet the requirement, the process proceeds to step S224; when there is no stable index period in which the coefficient of change of the associated index does not meet the requirement, the process proceeds to step S225; S224: when the number of stable time periods of the associated index whose variation coefficient does not meet the requirement is greater than the number of preset stable time periods, it is determined that the customer group does not belong to the surprise degree associated group; when the number of stable time periods of the associated index whose variation coefficient does not meet the requirement is not greater than the number of preset stable time periods, the process proceeds to step S225; S225 determines the comprehensive index variation coefficient based on the correlation index variation coefficients of different index stability periods and the time intervals between different index stability periods. When the comprehensive index variation coefficient does not meet the requirements, it is determined that the customer group does not belong to the surprise degree associated group. When the comprehensive index variation coefficient meets the requirements, it proceeds to step S23.

[0049] Furthermore, the composition data of the historical credit users in the index evaluation period includes the number of historical credit users in different surprise degree associated groups and the proportion of the number of historical credit users in the index evaluation period.

[0050] Specifically, Figure 4 As shown, the method for determining the explanation threshold of the surprise degree association group is: Determine the proportion of historical credit users in the indicator evaluation period based on the composition data of historical credit users corresponding to the surprise degree associated group in the indicator evaluation period; Determine the association weight coefficients of different surprise degree association groups according to the proportion of the number of historical credit users in different surprise degree association groups, and use the surprise degree association groups whose association weight coefficients are greater than the preset weight coefficients as the screening association groups; The deviation of the credit risk indicator in the indicator evaluation period from a preset benchmark risk indicator threshold is determined using the change data of the credit risk indicator in the indicator evaluation period, the deviation rate is determined using the ratio of the deviation amount to the preset benchmark risk indicator threshold, and the explanatory threshold of the surprise degree associated group is determined based on the number of the screened associated groups and the deviation rate.

[0051] Further, according to the number and deviation rate of the screened associated groups, the explanation threshold of the surprise associated groups is determined, specifically including: Determining a data matching coefficient based on the ratio of the number of the screened associated groups to the number of the customer groups; The compensation factor is determined according to the ratio of the deviation rate to the data matching coefficient, and the compensation threshold is determined according to the product of the preset interpretation threshold and the compensation factor; Based on the difference between the compensation threshold and a preset explanation threshold, the explanation threshold of the surprise degree associated group is determined.

[0052] Furthermore, a binary tree is constructed, including: Obtain the corresponding surprise-related index data, explanation threshold, and customer group proportion threshold; If the proportion of historical credit users in the surprise-related group is less than the customer group proportion threshold, the explanation and surprise of the customer group are calculated, and the customer group is used as a binary tree leaf node; If the proportion of historical credit users in the surprise-related group is not less than the customer group proportion threshold, Calculate the explanatory power and surprise power of the associated index of the customer group, and when the explanatory power of the surprise power associated group does not meet the constraint condition of the explanatory power threshold, use the surprise power associated group as a binary tree leaf node; When the explanatory power of the surprise degree associated group meets the constraint condition of the explanatory power threshold, the surprise degree associated group is secondary divided into subgroups based on the similarity of basic information of historical credit users of subgroups of the surprise degree associated group, and binary tree leaf nodes are constructed based on the subgroups whose surprise degrees meet the surprise degree threshold, and a decision tree is constructed using the binary tree leaf nodes.

[0053] Specifically, historical credit users whose basic information deviations in a certain dimension are within a preset deviation range are divided into the same sub-group.

[0054] It should be noted that the basic information includes occupation, income, region, credit limit, expenditure data and client.

[0055] Furthermore, the explanatory power of the association index of the surprise association group is determined based on the ratio of the deviation between the average value of the association index of the surprise association group in the evaluation target period and the association index in the stability index period, and the deviation between the average value of the association index of the credit risk indicator in the evaluation target period and the stability index period.

[0056] It should be noted that the method for determining the surprise degree of the surprise degree associated group is: Determine the benchmark values ​​in different unit periods based on the average values ​​of the credit risk indicators on different dates in different unit periods, and determine the indicator change dates on the dates based on the deviations between the credit risk indicators on different dates in the unit period and the benchmark values; The surprise degree of the customer group is determined according to the ratio of the number of indicator change dates in the evaluation target period to the number of indicator change dates in different indicator stable periods.

[0057] Furthermore, the attribution results of the credit risk indicators are generated and processed, including: The customer group whose explanatory degree of the binary tree leaf node meets the explanatory degree threshold is used as the output result of the credit risk indicator.

[0058] Example 1 In another possible embodiment, the method for determining the stable period of the credit risk indicator is: Determine the benchmark value in different unit periods by taking the average value of the credit risk indicator on different dates in different unit periods; Determine the deviation rate of different dates according to the deviation of the credit risk indicator on different dates in the unit period from the benchmark value and the ratio of the deviation to the benchmark value; Whether the unit period is a stable indicator period of the credit risk indicator is determined according to an average value of the deviation rates of different dates in the unit period.

[0059] Furthermore, when the average value of the deviation rates of different dates in the unit period is greater than a preset deviation rate threshold, it is determined that the unit period does not belong to a stable indicator period of the credit risk indicator.

[0060] Optionally, the method for determining the stable period of the credit risk indicator is: S11: determining the benchmark value in different unit periods by taking the average value of the credit risk indicator on different dates in different unit periods, and determining the benchmark deviation coefficient of the unit period by taking the deviation of the benchmark value between the unit period and other unit periods; S12: determining the deviation coefficients of different dates according to the deviations of the credit risk indicators of different dates in the unit period from the benchmark value and their ratios to the benchmark value, and combining the deviations of the credit risk indicators of different dates with those of adjacent dates, and determining the date change deviation coefficient of the unit period based on the deviation coefficients of different dates; S13 determines the indicator change amount of the unit period according to the average of the benchmark deviation coefficient of the unit period and the date variation deviation coefficient, and uses the indicator change amount to determine whether the unit period is an indicator stable period of the credit risk indicator.

[0061] Furthermore, when the indicator change amount of the unit time period is greater than a preset change amount threshold, it is determined that the unit time period does not belong to the indicator stable period of the credit risk indicator.

[0062] Optionally, the above step S11 includes the following contents: S111 determines the benchmark values ​​in different unit periods by taking the average values ​​of the credit risk indicators on different dates in different unit periods, and uses the deviation of the benchmark values ​​between the unit period and other unit periods to determine that when there are other unit periods in the unit period whose deviation of the benchmark value does not meet the requirements, the process proceeds to step S112; when there are no other unit periods in the unit period whose deviation of the benchmark value does not meet the requirements, the process proceeds to step S113; S112: when the deviation rate from the benchmark value of the unit period does not meet the requirement, other unit periods are taken as deviation unit periods; when the number of deviation unit periods is greater than the preset number of deviation unit periods, it is determined that the unit period does not belong to the indicator stability period of the credit risk indicator; when the number of deviation unit periods is not greater than the preset number of deviation unit periods, the process proceeds to step S113; S113 determines the benchmark deviation coefficient of the unit period by using the deviation of the benchmark value between the unit period and other unit periods. When the benchmark deviation coefficient of the unit period does not meet the requirements, it is determined that the unit period does not belong to the indicator stability period of the credit risk indicator. When the benchmark deviation coefficient of the unit period meets the requirements, proceed to step S12.

[0063] Optionally, the above step S12 includes the following contents: S121 determines the indicator change date in the unit period according to the deviation of the credit risk indicator on different dates in the unit period from the benchmark value. When there is no indicator change date in the unit period, it is determined that the unit period belongs to the indicator stable period of the credit risk indicator. When there is an indicator change date in the unit period, the process proceeds to step S122. S122: When the proportion of the number of indicator change dates in the unit period is greater than the proportion of the number of preset change dates, it is determined that the unit period does not belong to the indicator stability period of the credit risk indicator; when the proportion of the number of indicator change dates in the unit period is not greater than the proportion of the number of preset change dates, the process proceeds to step S123; S123: Determine the deviation coefficients of different dates according to the deviations of the credit risk indicators of different dates in the unit period from the benchmark value and the ratios to the benchmark value, and combine the deviations of the credit risk indicators of different dates with those of adjacent dates. When the average value of the deviation coefficients of different dates does not meet the requirements, determine that the unit period does not belong to the indicator stability period of the credit risk indicator. When the average value of the deviation coefficients of different dates meets the requirements, proceed to step S124. S124 determines the date variation deviation coefficient of the unit period based on the deviation coefficients of different dates. When the date variation deviation coefficient is greater than the preset variation deviation coefficient threshold, it is determined that the unit period does not belong to the indicator stability period of the credit risk indicator. When the date variation deviation coefficient is not greater than the preset variation deviation coefficient threshold, proceed to step S13.

[0064] Specifically, the associated indicator of the customer group is determined according to the type of the credit risk indicator. Specifically, when the credit risk indicator is the overdue rate, the associated indicator is the overdue data of different customers in the customer group.

[0065] Embodiment 2 Optionally, the method for determining the explanation threshold of the surprise degree associated group is: S31, determining the deviation of the credit risk indicator in the indicator evaluation period from a preset benchmark risk indicator threshold value based on the change data of the credit risk indicator in the indicator evaluation period, and determining the deviation rate based on the ratio of the deviation to the preset benchmark risk indicator threshold value; S32 determines the proportion of historical credit users in the indicator evaluation period based on the composition data of historical credit users corresponding to the surprise degree associated group in the indicator evaluation period, and determines the basic matching coefficient based on the proportion of historical credit users; S33 determines the association weight coefficients of different surprise degree association groups according to the proportion of the number of historical credit users in the different surprise degree association groups, and determines the data matching coefficient according to the association weight coefficients of the different surprise degree association groups and the proportion of the number of the surprise degree association groups in the customer group; S34 determines the explanation threshold of the surprise degree associated group according to the basic matching coefficient, the data matching coefficient and the deviation rate.

[0066] Further, according to the basic matching coefficient, the data matching coefficient and the deviation rate, the explanation threshold of the surprise degree associated group is determined, specifically including: The matching coefficient is determined by the sum of the basic matching times and the data matching coefficient, and the ratio of the deviation rate to the matching coefficient is used as the deviation matching factor; The preset explanation threshold corresponding to the deviation matching factor is used as the explanation threshold of the surprise degree group.

[0067] Optionally, before entering step S32, it is also necessary to determine whether the deviation rate is greater than a preset deviation rate threshold. If so, the explanatory threshold of the surprise group is determined using the set explanatory threshold. If not, enter step S32.

[0068] Optionally, the above step S32 includes the following contents: S321 determines the proportion of the number of historical credit users in the index evaluation period according to the composition data of the historical credit users corresponding to the surprise degree associated group in the index evaluation period, and determines the basic matching coefficient by using the proportion of the number of historical credit users. When the basic matching coefficient is less than the preset matching coefficient threshold, the explanation threshold of the surprise degree group is determined by using the set explanation threshold. When the basic matching coefficient is not less than the preset matching coefficient threshold, the process proceeds to step S322. S322: When the basic matching coefficient is within the preset matching coefficient interval, proceed to step S323; when the basic matching coefficient is within the preset matching coefficient interval, proceed to step S33; S323 When the number of historical credit users corresponding to the surprise degree associated group in the indicator evaluation period is less than the preset number of credit users, the explanation threshold of the surprise degree group is determined using the set explanation threshold; when the number of historical credit users corresponding to the surprise degree associated group in the indicator evaluation period is not less than the preset number of credit users, proceed to step S33.

[0069] Optionally, the above step S33 includes the following contents: S331 determines the association weight coefficients of different surprise association groups according to the proportion of the number of historical credit users in different surprise association groups. When the association weight coefficients of different surprise association groups are all less than the preset weight coefficients, the explanation threshold of the surprise group is determined by using the set explanation threshold. When there is a surprise association group whose association weight coefficient is not less than the preset weight coefficient, the process proceeds to step S332. S332 obtains the number of surprise degree associated groups whose associated weight coefficient is not less than the preset weight coefficient. When the number of surprise degree associated groups whose associated weight coefficient is not less than the preset weight coefficient is less than the preset associated group number threshold, the process proceeds to step S333. When the number of surprise degree associated groups whose associated weight coefficient is not less than the preset weight coefficient is not less than the preset associated group number threshold, the process proceeds to step S334. S333: When the proportion of the surprise degree associated group in the customer group is less than the proportion of the preset group, the explanation degree threshold of the surprise degree group is determined by using the set explanation degree threshold; when the proportion of the surprise degree associated group in the customer group is not less than the proportion of the preset group, the process proceeds to step S334; S334 determines the data matching coefficient based on the associated weight coefficients of different surprise-related groups and the proportion of the number of surprise-related groups in the customer group. When the data matching coefficient is less than the preset value of the matching coefficient, the explanatory threshold of the surprise group is determined using the set explanatory threshold. When the data matching coefficient is not less than the preset value of the matching coefficient, the process proceeds to step S34.

[0070] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0071] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The above description is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included in the scope of the claims of this specification.

Claims

1. A credit risk dynamic attribution analysis method based on binary tree, characterized in that: Specifically include: Use the credit product type to divide historical credit users into multiple customer groups, and determine the indicator stability period of the credit risk indicator based on the changes in the credit risk indicator in different unit time periods; Obtaining changes in the associated indicators of the customer groups in different indicator stability periods, and determining the surprise-related groups in the customer groups based on the changes in the associated indicators; Determine the composition data of the historical credit users corresponding to the surprise degree associated group during the index evaluation period, and determine the explanation threshold of the surprise degree associated group in combination with the number of the surprise degree associated group and the change data of the credit risk index during the index evaluation period; A binary tree is constructed based on the surprise degree associated groups and the explanation degree threshold, and binary tree leaf nodes and credit risk indicator attribution results are generated based on the explanation degrees and surprise degrees of different surprise degree associated groups.

2. The binary tree-based credit risk dynamic attribution analysis method according to claim 1, characterized in that: The credit product type is determined based on the credit products developed by the credit service platform for different groups.

3. The binary tree-based credit risk dynamic attribution analysis method according to claim 1, characterized in that: Divide historical credit users into multiple customer groups, including: Customers belonging to the same type of credit product are grouped into the same customer group.

4. The binary tree-based credit risk dynamic attribution analysis method according to claim 1, characterized in that: The credit risk indicators include the utilization rate, the overdue rate and the recovery rate.

5. The binary tree-based credit risk dynamic attribution analysis method according to claim 1, characterized in that: The method for determining the stable period of the credit risk indicator is: Determine the benchmark value in different unit periods by taking the average value of the credit risk indicator on different dates in different unit periods; Determine the index change date in the date according to the deviation between the credit risk index of different dates in the unit period and the benchmark value; Whether the unit period is a stable period of the credit risk indicator is determined based on the proportion of the number of dates on which the indicator changes.

6. The binary tree-based credit risk dynamic attribution analysis method according to claim 5, characterized in that: The indicator change date is the date when the deviation of the credit indicator from the benchmark value is not within the preset indicator deviation range.

7. The binary tree-based credit risk dynamic attribution analysis method according to claim 1, characterized in that: Construct a binary tree, including: Obtain the corresponding surprise-related index data, explanation threshold, and customer group proportion threshold; If the proportion of historical credit users in the surprise-related group is less than the customer group proportion threshold, the explanation and surprise of the customer group are calculated, and the customer group is used as a binary tree leaf node; If the proportion of historical credit users in the surprise-related group is not less than the customer group proportion threshold, Calculate the explanatory power and surprise power of the associated index of the customer group, and when the explanatory power of the surprise power associated group does not meet the constraint condition of the explanatory power threshold, use the surprise power associated group as a binary tree leaf node; When the explanatory power of the surprise degree associated group meets the constraint condition of the explanatory power threshold, the surprise degree associated group is secondary divided into subgroups based on the similarity of basic information of historical credit users of subgroups of the surprise degree associated group, and binary tree leaf nodes are constructed based on the subgroups whose surprise degrees meet the surprise degree threshold, and a decision tree is constructed using the binary tree leaf nodes.

8. The binary tree-based credit risk dynamic attribution analysis method according to claim 7, characterized in that: The basic information includes occupation, income, region, credit limit, spending data and client.

9. The binary tree-based credit risk dynamic attribution analysis method according to claim 7, characterized in that: The explanatory power of the association index of the surprise association group is determined based on the ratio of the deviation between the average value of the association index of the surprise association group in the evaluation target period and the association index in the stability index period, and the deviation between the average value of the association index of the credit risk indicator in the evaluation target period and the stability index period.

10. The binary tree-based credit risk dynamic attribution analysis method according to claim 7, characterized in that: Generate and process the attribution results of credit risk indicators, including: The customer group whose explanatory degree of the binary tree leaf node meets the explanatory degree threshold is used as the output result of the credit risk indicator.

Images