A predictive analysis method and platform for quality control

Abstract

The present invention provides a quality control prediction and analysis method and platform, which belongs to the field of prediction and evaluation technology, and specifically includes: evaluating the quality score of the credit model after the update processing in the prediction and analysis period, determining the model quality control method of the credit model based on the evaluation and analysis results of the quality score of the updated credit model in different prediction and analysis periods, determining the prediction and analysis method for quality control of the credit model after the next update processing based on the matching of the prediction and analysis periods between different update processing times of the credit model, and combining the model quality control method of the credit model at different update processing times, thereby improving the model quality of the credit model after the update processing.

Description

Technical Field

[0001] The present invention belongs to the technical field of prediction and evaluation, and in particular relates to a prediction and analysis method and platform for quality control. Background Art

[0002] In order to evaluate credit risk, credit companies often evaluate users' credit risk by establishing credit models. However, at the same time, after modeling, there are often deviations in the reliability of risk identification in the verification set. Therefore, in order to achieve quality control of the credit model, the existing technical solutions often perform quality control by setting a threshold corresponding to a fixed loss function. Specifically, a similar technical solution is given in the invention patent application CN202510329033.9 "A product recommendation method for multimodal data".

[0003] Using a fixed threshold to control the quality of the credit model may lead to the overlap of customer groups with which the original credit model has a deviation in recognition reliability as the customer group faced by the credit model changes, making it difficult for the credit risk control reliability based on the credit model to meet the requirements. Therefore, it is often necessary to implement differentiated quality control measures for the credit model based on the application scenario of the credit model and the changes in the credit risk recognition and processing quality of the credit model during the update process. However, at the same time, how to determine the analysis and comparison objects of the credit model after the update process, and ensure that the accuracy of the stability of the model's evaluation processing meets the requirements on the basis of improving the efficiency of the evaluation and analysis processing, has become a technical problem that needs to be solved urgently.

[0004] To solve the above technical problems, this application provides a quality control prediction analysis method and platform. Summary of the Invention

[0005] To achieve the purpose of the present invention, the present invention adopts the following technical solutions:

[0006] Specifically, this application provides a quality control prediction analysis method, which specifically includes:

[0007] S1 updates the credit model based on historical credit application data. When the composition of the verification deviation group of the credit model is determined to meet the requirements based on the verification data set of the credit model, the next step is entered;

[0008] S2 determines the forecast analysis period for quality control of the updated credit model based on the matching of the verification data of the updated credit model in the verification data set with the customer groups of the credit analysis period corresponding to the original different credit models;

[0009] S3: evaluating the quality score of the updated credit model in the forecast analysis period, and determining a method for controlling the model quality of the credit model based on evaluation analysis results of the quality scores of the updated credit model in different forecast analysis periods;

[0010] S4 determines the prediction analysis method for quality control of the credit model after the next update process based on the matching of the prediction analysis periods between different update processing times of the credit model and the control method of the model quality of the credit model at different update processing times.

[0011] The beneficial effects of the present invention are:

[0012] The control method of the model quality of the credit model is determined based on the evaluation and analysis results of the quality scores of the updated credit model in different forecast analysis periods, thereby realizing the determination of the quality score based on the recognition deviation of different credit users between the original credit model and the updated credit model, and then realizing the evaluation of the stability of the credit risk identification processing of the credit model from the perspective of the quality score, thereby realizing the determination of the control method of the model quality of the credit model after the next update from the perspective of the stability, that is, ensuring the efficiency of the training processing of the credit model with a higher degree of stability, and also ensuring the reliability of the training processing results of the credit model with a lower degree of stability.

[0013] Based on the matching of the prediction and analysis time periods of the credit model between different update processing times and the model quality control method of the credit model at different update processing times, the prediction and analysis method used for quality control of the credit model after the next update processing is determined, thereby avoiding the technical problem of low stability and reliability of the identification processing of the credit risk of the credit model caused by using the same prediction and analysis time period in different update processing processes. By further combining the model quality control method, the reliability of the model quality control method and the distribution and aggregation degree of the prediction and analysis time period are realized, and the prediction and analysis time period of the credit model is determined in the next update processing of the credit model, thereby ensuring the reliability of the evaluation processing of the stability of the identification processing of the credit risk of the credit model.

[0014] A further technical solution is that the historical credit application data includes credit application data of different credit application users.

[0015] It should be noted that the credit application data includes one or more of the applicant's income certificate, bank statements, credit report, real estate certificate, vehicle driving license, marriage certificate, academic certificate, and tax payment certificate, and all need to be collected with the user's consent.

[0016] A further technical solution is that the credit model is updated according to a preset time period, and specifically a training data set and a verification data set are constructed using credit application data of different credit application users to perform update training of the credit model.

[0017] It is understandable that the credit model is constructed through one or more of logistic regression, decision tree, random forest, support vector machine, and neural network.

[0018] A further technical solution is that the verification deviation group is determined by the recognition accuracy of credit risk in different customer groups in the verification data set, and is specifically determined based on the proportion of successfully identified credit risk users in the credit risk users in the verification data set, and the customer group with an identification processing rate less than 0.9 is used as the verification deviation group.

[0019] A further technical solution is that the method for determining the predictive analysis method for quality control of the credit model after the next update process is:

[0020] Based on the control method of the credit model with different update processing times, determine the update processing times of the control method using different model quality;

[0021] Based on the matching of the forecast analysis periods of the credit model between different update processing times, the number of different credit analysis periods as the forecast analysis period is determined and used as the forecast analysis quantity;

[0022] The number of forecast analyses in different credit analysis periods and the number of update processing times of different model quality control methods are used to determine the forecast analysis method used for quality control of the credit model after the next update processing.

[0023] In a second aspect, the present application provides a quality control prediction and analysis platform, which adopts the above-mentioned quality control prediction method, specifically including:

[0024] Analysis period identification module, quality control module, analysis strategy determination module;

[0025] The analysis period identification module is responsible for determining the forecast analysis period of the updated credit model for quality control;

[0026] The quality control module is responsible for determining a control method for the model quality of the credit model;

[0027] The analysis strategy determination module is responsible for determining the predictive analysis method used for quality control of the credit model after the next update process.

[0028] Other features and advantages will be described in the following description. The objectives and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and drawings.

[0029] 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

[0030] 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 accompanying drawings.

[0031] Figure 1 It is a flow chart of a predictive analysis method for quality control;

[0032] Figure 2 It is a flow chart to determine whether the composition of the validation deviation group of the credit model meets the requirements;

[0033] Figure 3 is a flow chart of a method for determining a forecast analysis period for quality control of a credit model;

[0034] Figure 4 is a flow chart of a method for determining a predictive analysis method for quality control of a credit model after a next update process;

[0035] Figure 5 It is a framework diagram of a predictive analysis platform for quality control. DETAILED DESCRIPTION

[0036] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this specification without creative work should fall within the scope of protection of this specification.

[0037] In this application, by combining the credit risk recognition stability of the credit model in the most recent six months or one year, the differentiated model quality control of the credit model during the update and training process is achieved, thereby improving the model quality, and targeted optimization is performed on the evaluation object of the model's recognition stability, further improving the efficiency and reliability of the model's recognition stability.

[0038] The verification deviation group is a customer group whose credit risk identification accuracy is less than 0.9. When the number of verification deviation groups is greater than the preset deviation group number threshold, it is determined that the composition of the verification deviation group of the credit model does not meet the requirements.

[0039] The credit analysis period corresponding to the credit model is the period between the moment when the credit model begins to be used for credit risk assessment and analysis and the moment when the next updated credit model begins to be used for credit risk assessment and analysis. The original different credit models are credit models updated within the most recent preset period. In a possible embodiment, they are credit models updated within the most recent six months or one year.

[0040] Based on the verification data of the verification data set, the recognition accuracy in different customer groups is determined. If the recognition accuracy of the updated credit model in different customer groups is greater than that of the credit model, the credit analysis period corresponding to the credit model will be used as the preset analysis period.

[0041] The quality score is determined based on the proportion of customers whose credit assessment processing results of different customers of the updated credit model in different credit analysis periods are inconsistent with the credit assessment processing results of customers of the original credit model corresponding to the credit analysis period. In a possible embodiment, the quality score is determined based on the difference between 1 and the proportion of the number of customers.

[0042] When the average value of the quality scores of different prediction analysis periods is less than the preset quality score threshold, it means that the stability of the credit model is not high. Therefore, on this basis, when controlling the model quality of the credit model during the next update process, the first control strategy is adopted, that is, on the basis of controlling the overall prediction accuracy to be greater than the preset accuracy, it is necessary to further combine the control of the model quality of the updated credit model by controlling the number of verification deviation groups. Specifically, the number of verification deviation groups is controlled within 2 and the proportion of the number of users of the verification deviation group in the verification data set is controlled within 0.05.

[0043] If the average value of the quality analysis in different forecast analysis periods is greater than the preset quality score threshold, the second control strategy is adopted. This indicates that the credit model is relatively stable. Therefore, on this basis, it is only necessary to control the overall forecast accuracy to be greater than the preset accuracy, that is, greater than 0.98.

[0044] When the number of credit analysis periods used as forecast analysis periods is less than the preset analysis number threshold and does not meet the requirement, or when the number of credit analysis periods with forecast analysis times less than 3 is more than 5, then since the number of periods for stability analysis of the credit model is small, in order to ensure the reliability of quality control, the forecast analysis method used for quality control after the next credit model update is to use all credit analysis periods as forecast analysis periods;

[0045] When the number of credit analysis periods with a forecast analysis number less than the preset analysis number threshold meets the requirements, when the number of update processing times for model quality control using the second control strategy does not meet the requirements, that is, the proportion of the number of update processing times in the most recent preset time period is greater than 0.33, then in order to ensure the reliability of quality control, the forecast analysis method used for quality control after the next update processing of the credit model is to use all credit analysis periods as forecast analysis periods.

[0046] Example 1

[0047] like Figure 1 As shown, the present application provides a predictive analysis method for quality control, specifically comprising:

[0048] S1 updates the credit model based on historical credit application data. When the composition of the verification deviation group of the credit model is determined to meet the requirements based on the verification data set of the credit model, the next step is entered;

[0049] Furthermore, the historical credit application data includes credit application data of different credit application users.

[0050] It should be noted that the credit application data includes one or more of the applicant's income certificate, bank statements, credit report, real estate certificate, vehicle driving license, marriage certificate, academic certificate, and tax payment certificate, and all need to be collected with the user's consent.

[0051] Specifically, the credit model is updated according to a preset time period. Specifically, a training data set and a verification data set are constructed using credit application data of different credit application users to perform update training of the credit model.

[0052] It is understandable that the credit model is constructed through one or more of logistic regression, decision tree, random forest, support vector machine, and neural network.

[0053] Furthermore, the verification deviation group is determined by the recognition accuracy of credit risk in different customer groups in the verification data set, and is specifically determined based on the proportion of successfully identified credit risk users in the credit risk users in the verification data set. The customer group with an identification processing rate less than 0.9 is taken as the verification deviation group. It can be understood that the credit risk users are users with credit risks, that is, users who do not meet the credit conditions.

[0054] Furthermore, the customer groups are divided according to the same number in different credit application data. Specifically, the credit application users whose same number of credit application data meets the requirements are divided into the same customer group. In a possible embodiment, the credit application users whose same number of credit application data accounts for more than 0.6 of the number in the credit application data are divided into the same customer group.

[0055] Specifically, such as Figure 2 As shown, it is determined that the composition of the verification deviation group of the credit model meets the requirements, specifically including:

[0056] Determining the number of verification deviation groups in the credit model based on the composition of the verification deviation groups in the credit model;

[0057] Determine the proportion of credit application users in the verification deviation group in the verification data set based on the number of credit application users in the verification deviation group in the credit model, and use it as the deviation user proportion;

[0058] The proportion of deviation users is used to determine whether the composition of the verification deviation group of the credit model meets the requirements.

[0059] It can be understood that when the proportion of deviation users is greater than the preset deviation user proportion threshold, it is determined that the composition of the verification deviation group of the credit model does not meet the requirements.

[0060] Furthermore, when the composition of the verification deviation group of the credit model does not meet the requirements, quality score evaluation and analysis are performed in the credit analysis periods corresponding to different original credit models, and the quality scores in the credit analysis periods are used to determine the control method for the model quality of the credit model during the next update processing.

[0061] It should be noted that the credit analysis period corresponding to the credit model is the period between the moment when the credit model starts to be used for credit risk assessment and analysis and the moment when the credit model after the next update starts to be used for credit risk assessment and analysis.

[0062] Furthermore, the original different credit models are credit models that have been updated within the most recent preset period, and in a possible embodiment, are credit models that have been updated within the most recent six months or one year.

[0063] Optionally, determining whether the composition of the verification deviation group of the credit model meets the requirements specifically includes:

[0064] Determining the number of verification deviation groups in the credit model based on the composition of the verification deviation groups in the credit model;

[0065] According to the number of verification deviation groups in the credit model, it is determined whether the composition of the verification deviation groups of the credit model meets the requirements.

[0066] Furthermore, when the number of verification deviation groups is greater than a preset deviation group number threshold, it is determined that the composition of the verification deviation groups of the credit model does not meet the requirements.

[0067] S2 determines the forecast analysis period for quality control of the updated credit model based on the matching of the verification data of the updated credit model in the verification data set with the customer groups of the credit analysis period corresponding to the original different credit models;

[0068] Furthermore, the verification data is the recognition accuracy of credit risk in different customer groups in the verification data set.

[0069] Specifically, the original different credit models are different updated credit models in history.

[0070] In a possible embodiment, the credit model is updated every month, and thus a credit model exists in every month.

[0071] Specifically, such as Figure 3 As shown, the method for determining the forecast analysis period for quality control using the credit model is as follows:

[0072] Using the updated credit model as the updated credit model, and based on the validation data of the credit model in the validation data set, determining the credit risk recognition accuracy of the updated credit model for different customer groups in the validation data set;

[0073] Determining a reliable group for identification changes based on a deviation between the identification processing accuracy rate and the identification processing accuracy rate of the original credit model;

[0074] Whether the credit analysis period is a forecast analysis period is determined based on the distribution data of the identified change reliable group in the credit analysis period corresponding to the original credit model.

[0075] Optionally, the identification change reliable group is a customer group whose identification processing accuracy in the updated credit model is greater than the identification processing accuracy of the original credit model.

[0076] Furthermore, when the number of credit application users in the identified change reliable group in the credit analysis period corresponding to the original credit model is greater than a preset user number threshold, the credit analysis period is determined to be a forecast analysis period.

[0077] In another embodiment, when, in the credit analysis period corresponding to the original credit model, there are credit application users whose identification changes in the reliable group on different dates, the credit analysis period is determined to be a forecast analysis period.

[0078] Optionally, the method for determining the forecast analysis period for quality control of the credit model is:

[0079] S21 uses the updated credit model as an updated credit model, and based on the verification data of the updated credit model in the verification data set, determines the recognition accuracy of the updated credit model for credit risks in different customer groups in the verification data set, and determines the reliable customer group based on the recognition accuracy;

[0080] It can be understood that in the above steps, if the average recognition accuracy of the updated credit model for credit risk in different customer groups in the verification data set is greater than the preset recognition accuracy threshold, then since the recognition accuracy of the updated credit model is higher, the credit analysis period corresponding to the different original credit models is used as the predictive analysis period for quality control of the credit model, thereby facilitating a more accurate assessment of the changing trend of the quality score of the credit model.

[0081] It should also be noted that even if the average of the credit risk identification accuracy rates of the updated credit model in different customer groups in the validation data set is not greater than the preset identification accuracy threshold, the customer group with an identification accuracy rate greater than the preset accuracy threshold is used as the identified reliable customer group. In one embodiment, the customer group with an identification accuracy rate greater than 0.98 is used as the identified reliable customer group.

[0082] Specifically, when the number of identified reliable customer groups meets the requirement, in one possible embodiment, when the number of identified reliable customer groups in the customer group accounts for more than 0.7, the credit analysis period corresponding to the different original credit models is used as the forecast analysis period for the credit model for quality control, thereby facilitating more accurate assessment of the changing trend of the credit model's quality score;

[0083] It should also be noted that, if the proportion of the number of identified reliable customer groups in the customer group is not greater than 0.7, the identification reliability value is determined based on the average of the proportion of the number of identified reliable customer groups in the customer group and the average of the identification accuracy rates of credit risks in different customer groups. When the identification reliability value is greater than the preset reliability threshold, the credit analysis periods corresponding to different original credit models are used as the forecast analysis periods for quality control of the credit models, thereby facilitating a more accurate assessment of the changing trend of the quality score of the credit model.

[0084] S22 determines a reliable group for identification changes based on a deviation between the identification processing accuracy rate and the identification processing accuracy rate of the original credit model;

[0085] It should also be noted that, in one possible embodiment, if the number of identified change reliable groups meets the requirement, that is, in one possible embodiment, the number of identified change reliable groups accounts for more than 0.3 of the total number of customer groups, then the credit analysis periods corresponding to different original credit models are used as the forecast analysis periods for quality control of the credit models, thereby facilitating more accurate assessment of the changing trend of the quality score of the credit models;

[0086] S23 determines whether the credit analysis period is a predictive analysis period based on the distribution data of the identified changing reliable groups and the identified reliable customer groups in the credit analysis period corresponding to the original credit model, and combined with the deviation between the identification accuracy of credit risks in different customer groups and the identification accuracy of the original credit model.

[0087] It can be understood that in a possible embodiment, the deviation between the recognition accuracy of credit risk in different customer groups and the recognition accuracy of the original credit model is used as a weight value to identify the proportion of the number of credit application users of the reliable customer group in the credit analysis period corresponding to the original credit model, identify the weight of the proportion of the number of credit application users of the changing reliable group in the credit analysis period corresponding to the original credit model, and determine the analysis matching value. When the analysis matching value is greater than the preset matching threshold, the credit analysis period is determined to be a predictive analysis period.

[0088] Specifically, the quality score is determined based on the proportion of customers whose credit assessment processing results of different customers in different credit analysis periods under the updated credit model are inconsistent with the credit assessment processing results of customers of the original credit model corresponding to the credit analysis period. In a possible embodiment, the quality score is determined based on the difference between 1 and the proportion of the number of customers.

[0089] S3: evaluating the quality score of the updated credit model in the forecast analysis period, and determining a method for controlling the model quality of the credit model based on evaluation analysis results of the quality scores of the updated credit model in different forecast analysis periods;

[0090] Furthermore, after the updated credit model obtains quality scores in different forecast analysis periods, the average value of the quality scores in different forecast analysis periods is used to determine the control method for the model quality of the credit model during the next update process.

[0091] Specifically, when the average value of the quality scores of different prediction analysis periods is less than the preset quality score threshold, it means that the stability of the credit model is not high. Therefore, on this basis, when controlling the model quality of the credit model during the next update process, the first control strategy is adopted, that is, on the basis of controlling the overall prediction accuracy to be greater than the preset accuracy, it is necessary to further combine the control of the model quality of the updated credit model by controlling the number of verification deviation groups. Specifically, the number of verification deviation groups is controlled within 2 and the proportion of the number of users of the verification deviation group in the verification data set is controlled within 0.05.

[0092] If the average value of the quality analysis in different forecast analysis periods is greater than the preset quality score threshold, the second control strategy is adopted. This indicates that the credit model is relatively stable. Therefore, on this basis, it is only necessary to control the overall forecast accuracy to be greater than the preset accuracy, that is, greater than 0.98.

[0093] S4 determines the prediction analysis method for quality control of the credit model after the next update process based on the matching of the prediction analysis periods between different update processing times of the credit model and the control method of the model quality of the credit model at different update processing times.

[0094] Specifically, such as Figure 4 As shown, the method for determining the predictive analysis method for quality control of the credit model after the next update process is:

[0095] Based on the control method of the credit model with different update processing times, determine the update processing times of the control method using different model quality;

[0096] Based on the matching of the forecast analysis periods of the credit model between different update processing times, the number of different credit analysis periods as the forecast analysis period is determined and used as the forecast analysis quantity;

[0097] The number of forecast analyses in different credit analysis periods and the number of update processing times of different model quality control methods are used to determine the forecast analysis method used for quality control of the credit model after the next update processing.

[0098] Specifically, when the number of credit analysis periods with a forecast analysis number less than a preset analysis number threshold does not meet the requirement, or when the number of credit analysis periods with a forecast analysis number less than three is more than five, then since the number of periods for performing stability analysis on the credit model is small, in order to ensure the reliability of quality control, the forecast analysis method used for quality control after the next update of the credit model is to use all credit analysis periods as forecast analysis periods;

[0099] It should also be noted that when the number of credit analysis periods with a forecast analysis number less than the preset analysis number threshold meets the requirements, it is necessary to further determine the number of update processing times for model quality control using the second control strategy. When the number of update processing times for model quality control using the second control strategy does not meet the requirements, that is, the proportion of the number of update processing times in the most recent preset time period is greater than 0.33, then in order to ensure the reliability of quality control, the forecast analysis method used for quality control after the next update of the credit model is to use all credit analysis periods as forecast analysis periods.

[0100] In other cases, the prediction analysis period for quality control of the updated credit model is determined based on the matching of the verification data of the updated credit model in the verification data set with the customer groups in the credit analysis period corresponding to the original different credit models.

[0101] Optionally, the method for determining the predictive analysis method for quality control of the credit model after the next update process is:

[0102] Based on the control methods of the credit model with different update processing times, determining the update processing times for using different model quality control methods, determining the proportion of update processing times for model quality control using the second control strategy based on the update processing times, and using this proportion as the secondary update times proportion, determining the model quality control deviation value based on the secondary update times proportion and the model quality control method of the credit model with the most recent preset times threshold;

[0103] It should be noted that, in the above steps, if the control method for the model quality of the credit model after the next update process is the second control strategy, all credit analysis periods will be used as forecast analysis periods.

[0104] It is also understandable that, if the control method for the model quality of the credit model after the next update process is the first control strategy in the above steps, it is determined whether the proportion of the number of secondary updates meets the requirements. When the proportion of the number of secondary updates is greater than 0.33, in order to ensure the reliability of quality control, the predictive analysis method used for quality control after the next update process of the credit model is to use all credit analysis periods as predictive analysis periods;

[0105] Furthermore, even if the proportion of the second update times is not greater than 0.33, it is necessary to further combine the model quality control method of the credit model with the most recent preset number threshold. It can be understood that when the model quality control method of the credit model for the last three times has more than two update processing times using the second control strategy, in order to ensure the reliability of quality control, the predictive analysis method used for quality control after the next update processing of the credit model is to use all credit analysis periods as predictive analysis periods.

[0106] In other cases, in a possible embodiment, the model quality control method of the credit model based on the proportion of the second update times and the most recent preset number threshold is to multiply the proportion of the number of models of the second control strategy to determine the control deviation value. When the control deviation value is greater than the preset control deviation threshold, in order to ensure the reliability of quality control, the predictive analysis method used for quality control after the next update of the credit model is to use all credit analysis periods as predictive analysis periods. In other cases, proceed to the next step.

[0107] Determining different credit analysis periods as the number of forecast analysis periods based on matching conditions of the forecast analysis periods between different update processing times of the credit model, and using the different credit analysis periods as the forecast analysis quantity, and determining the number of credit analysis periods whose forecast analysis quantity is less than a preset analysis quantity threshold based on the forecast analysis quantity;

[0108] Furthermore, in the above steps, if the number of credit analysis periods in which the number of forecast analyses is less than the preset analysis number threshold does not meet the requirements, when the number of credit analysis periods in which the number of forecast analyses is less than 3 times is more than 5, then since the number of periods for stability analysis of the credit model is small, in order to ensure the reliability of quality control, the forecast analysis method used for quality control after the next update of the credit model is to use all credit analysis periods as forecast analysis periods, and in other cases, proceed to the next step.

[0109] The number of credit analysis periods in which the number of predicted analyses is less than a preset analysis number threshold and the control deviation value of the model quality are used to determine the predicted analysis method used for quality control after the next update process of the credit model.

[0110] Specifically, in one possible embodiment, a preset credit analysis period number threshold under the control deviation value is determined based on the control deviation value of the model quality; when the number of credit analysis periods in which the number of predicted analyses is less than the preset analysis number threshold is greater than the preset credit analysis period number threshold, it is determined that the prediction analysis method used for quality control of the credit model after the next update process is to use all credit analysis periods as prediction analysis periods;

[0111] In other cases, the prediction analysis period for quality control of the updated credit model is determined based on the verification data of the updated credit model in the verification data set and the matching of the customer groups in the credit analysis period corresponding to the original different credit models.

[0112] Example 2

[0113] Second, as Figure 5 As shown, the present application provides a quality control prediction and analysis platform, which adopts the above-mentioned quality control prediction method, specifically including:

[0114] Analysis period identification module, quality control module, analysis strategy determination module;

[0115] The analysis period identification module is responsible for determining the forecast analysis period of the updated credit model for quality control;

[0116] The quality control module is responsible for determining a control method for the model quality of the credit model;

[0117] The analysis strategy determination module is responsible for determining the predictive analysis method used for quality control of the credit model after the next update process.

[0118] The various embodiments in this specification are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are relatively simplified. For relevant details, refer to the descriptions of the method embodiments.

[0119] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0120] The foregoing description is merely one or more embodiments of this specification and is not intended to limit this specification. It will be apparent to those skilled in the art that various modifications and variations may be made to one or more embodiments of this specification. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of one or more embodiments of this specification are intended to be within the scope of the claims of this specification.

Claims

1. A predictive analysis method for quality control, characterized in that: Specifically include: The credit model is updated based on the historical credit application data. When the composition of the verification deviation group of the credit model is determined to meet the requirements based on the verification data set of the credit model, the next step is entered; Determining a forecast analysis period for quality control of the updated credit model based on the matching of the validation data of the updated credit model in the validation data set with the customer groups of the credit analysis period corresponding to the original different credit models; Performing an evaluation process on the quality score of the updated credit model during the forecast analysis period, and determining a method for controlling the model quality of the credit model based on evaluation analysis results of the quality score of the updated credit model during different forecast analysis periods; Determining the prediction analysis method for quality control of the credit model after the next update process based on the matching of the prediction analysis periods between different update process times and the model quality control method of the credit model at different update process times; The method for determining the forecast analysis period for quality control of the credit model is as follows: Using the updated credit model as the updated credit model, and based on the validation data of the credit model in the validation data set, determining the credit risk recognition accuracy of the updated credit model for different customer groups in the validation data set; Determining a reliable group for identification changes based on a deviation between the identification processing accuracy rate and the identification processing accuracy rate of the original credit model; determining whether the credit analysis period is a forecast analysis period based on distribution data of the identified change reliable group in the credit analysis period corresponding to the original credit model; The identification change reliable group is a customer group whose identification processing accuracy in the updated credit model is greater than that of the original credit model; The verification deviation group is determined by the recognition accuracy rate of credit risk in different customer groups in the verification data set, specifically determined according to the proportion of successfully identified credit risk users in the verification data set among credit risk users, and the customer group with an recognition processing rate less than 0.9 is regarded as the verification deviation group; The quality score is determined based on the proportion of customers whose credit assessment processing results of different customers in different credit analysis periods under the updated credit model are inconsistent with the credit assessment processing results of customers of the original credit model corresponding to the credit analysis period, and is determined based on the difference between 1 and the proportion of the number of customers.

2. The quality control prediction analysis method according to claim 1, characterized in that: The historical credit application data includes credit application data of different credit application users.

3. The quality control prediction analysis method according to claim 1, characterized in that: The credit model is updated according to a preset time period. Specifically, a training data set and a verification data set are constructed using credit application data of different credit application users to perform update training of the credit model.

4. The predictive analysis method for quality control according to claim 1, wherein: The validation deviation group is used to determine the recognition accuracy of credit risk in different customer groups in the validation data set.

5. The predictive analysis method for quality control according to claim 1, wherein: Determine that the composition of the validation bias group for the credit model meets the requirements, including: Determining the number of verification deviation groups in the credit model based on the composition of the verification deviation groups in the credit model; Determine the proportion of credit application users in the verification deviation group in the verification data set based on the number of credit application users in the verification deviation group in the credit model, and use it as the deviation user proportion; The proportion of deviation users is used to determine whether the composition of the verification deviation group of the credit model meets the requirements.

6. The predictive analysis method for quality control according to claim 1, wherein: The verification data is the recognition accuracy of credit risk in different customer groups in the verification data set.

7. The predictive analysis method for quality control according to claim 1, wherein: The method for determining the predictive analysis method used for quality control of the credit model after the next update process is: Based on the control method of the credit model with different update processing times, determine the update processing times of the control method using different model quality; Based on the matching of the forecast analysis periods of the credit model between different update processing times, the number of different credit analysis periods as the forecast analysis period is determined and used as the forecast analysis quantity; The number of forecast analyses in different credit analysis periods and the number of update processing times of different model quality control methods are used to determine the forecast analysis method used for quality control of the credit model after the next update processing.

8. The predictive analysis method for quality control according to claim 6, wherein: When the number of credit analysis periods with a forecast analysis quantity less than a preset analysis quantity threshold does not meet the requirement, the forecast analysis method used for quality control after the next update of the credit model is to use all credit analysis periods as forecast analysis periods.

9. A quality control prediction analysis platform, using a quality control prediction method according to any one of claims 1 to 8, characterized in that: Specifically include: Analysis period identification module, quality control module, analysis strategy determination module; The analysis period identification module is responsible for determining the forecast analysis period of the updated credit model for quality control; The quality control module is responsible for determining a control method for the model quality of the credit model; The analysis strategy determination module is responsible for determining the predictive analysis method used for quality control of the credit model after the next update process.

Images