Business risk prediction method and device, equipment, storage medium and program product

By optimizing the business risk prediction model through clustering and distributed gradient boosting algorithms, the problems of weak cross-scenario capabilities and high complexity in existing technologies are solved, and efficient business risk prediction across scenarios is achieved.

CN118940906APending Publication Date: 2024-11-12SHANGHAI HUARUI BANK CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411072450.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-06
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

Existing technologies cannot effectively manage short-term, medium-term and long-term risks simultaneously in business risk prediction, have weak cross-scenario capabilities, and have high model deployment complexity and high resource consumption.

Method used

By acquiring and grouping training samples, an initial prediction model is constructed using a distributed gradient boosting algorithm. The discrimination and hyperparameters are calculated, the objective function is optimized to determine the optimal parameters, and the model is adjusted to adapt to different risk types, thereby achieving cross-scenario risk prediction.

Benefits of technology

It has achieved good business risk prediction results in various scenarios, including short-term, medium-term and long-term, reduced the complexity of model deployment and resource consumption, enhanced cross-scenario capabilities, and improved risk identification and early warning capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940906A_ABST
    Figure CN118940906A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of risk control, discloses a business risk prediction method, device and equipment, a storage medium and a program product, and is used for solving the technical problems that the risk prediction effect is poor and the cross-scene capability is weak under various conditions of short term, middle term and long term in the prior art. The method comprises the following steps: acquiring a training subset and a test set; weight parameters are given to a loss function of the training subset, an initial prediction model is constructed, the distinction degree of the initial prediction model at all weight numerical value samples and model hyper-parameter samples on a test set is calculated, and a weight search interval and a model hyper-parameter search interval are determined; constructing an optimization objective function based on the weight search interval, the model hyper-parameter search interval and the distinction degree, performing optimization calculation to obtain an optimal weight parameter and a model hyper-parameter, performing parameter adjustment on the initial prediction model, and fitting a training set to obtain a trained risk prediction model; and performing business risk prediction on the to-be-predicted data based on the risk prediction model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of risk control technology, and in particular to a business risk prediction method, device, equipment, storage medium and program product. Background Art

[0002] At present, when predicting business risks, it is generally necessary to jointly manage risks in different scenarios. For example, in the consumer finance industry, in order to improve the ability to predict overdue risks of businesses with different ages, the existing technology adopts a variety of different solutions to achieve joint management of short-term and long-term risks; generally including the following solutions:

[0003] (1) Cross-application: Use both short-term and long-term models to predict risks and obtain two overdue probability scores. According to business needs and model performance, set decision rules to integrate the prediction results of the two models. However, this solution consumes a lot of resources and is complex to deploy.

[0004] (2) Transfer learning: By using a model that has been trained on one task, it can help improve the performance of another related task. However, this solution depends on the effectiveness of the transferred model, the transfer effect is difficult to guarantee, and additional model adjustment is required.

[0005] (3) Time series analysis: Using time series analysis methods and deep learning models such as long short-term memory networks, long-term and short-term risk factors can be captured in one model; however, this model is relatively complex and parameter adjustment is cumbersome, the model is relatively weak in interpretability, and is highly dependent on data;

[0006] (4) Ensemble learning method: By combining the prediction results of multiple models, the ensemble learning method can improve the robustness and accuracy of the model; however, this solution has high computational overhead and the model may not accurately predict the results of individual tasks.

[0007] In summary, there is an urgent need for a business risk prediction method that can simultaneously manage short-term, medium-term, and long-term business risks and improve the prediction effect, so that the business risk prediction method can have better cross-scenario capabilities, while simplifying the complexity of model deployment and reducing resource consumption. Summary of the invention

[0008] The main purpose of the present invention is to solve the technical problems in the prior art that when predicting business risks, the prediction effect cannot be guaranteed in various situations in the short, medium and long term, and the cross-scenario ability of the business risk prediction method is weak.

[0009] A first aspect of the present invention provides a business risk prediction method, comprising:

[0010] Acquire multiple training samples, and divide the training samples into a training set and a test set, and based on the numerical characteristics of each of the training samples, perform sample grouping on the training set according to a sample grouping rule to obtain multiple training subsets;

[0011] Assign weight parameters to the loss functions of multiple training subsets according to the number of training subsets, and call the distributed gradient boosting algorithm to build an initial prediction model;

[0012] Obtain multiple weight value samples and model hyperparameter samples, calculate the discrimination of the initial prediction model at each weight value sample and the discrimination of the model hyperparameter sample, and determine the weight search interval and the model hyperparameter search interval based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of short-term, medium-term and long-term;

[0013] Based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, an initial optimization objective function is constructed and an optimization calculation is performed to obtain the optimal weight parameter and the optimal model hyperparameter of each training subset;

[0014] Adjusting the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters, and obtaining a trained risk prediction model after fitting the training set;

[0015] Receive and respond to business risk prediction requests to obtain data to be predicted; input the data to be predicted into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted.

[0016] Optionally, in a first implementation of the first aspect of the present invention, the distributed gradient boosting algorithm is an XGBoost algorithm, and the weighted objective function expression included in the initial prediction model is:

[0017]

[0018] Where obj(θ) represents the weighted objective function of the prediction model; θ represents the model parameters; m represents the total number of training subsets; ω j is the weight parameter of the jth training subset; n j is the number of samples in the jth training subset, where j = 1, 2, 3, …, m; y i Represents the true value of the i-th training sample; represents the value of the i-th training sample predicted by the model based on the input; ω(f k ) represents the complexity of the kth subtree, and K is the total number of subtrees.

[0019] Optionally, in a second implementation of the first aspect of the present invention, the numerical features of the training samples include the number of business periods and the number of business overdue days;

[0020] Based on the numerical features of each training sample, the training set is grouped according to the sample grouping rule to obtain multiple training subsets including:

[0021] The training samples in the training set are grouped and labeled according to the business period and overdue days. If there are no business overdue days in the first 6 business periods, they are labeled as long-term good users;

[0022] If the number of overdue days in the first 6 business periods is greater than or equal to 30, the user will be marked as a long-term bad user;

[0023] Among the long-term bad users, find the training samples that are overdue in the first business period and the number of business overdue days is greater than or equal to 15, mark them as short-term bad users, and obtain the first training subset;

[0024] Among the long-term bad users, find the training samples that are overdue in the first business period and whose business overdue days are greater than or equal to 1 and less than 15, mark them as short-term gray users, and obtain the second training subset;

[0025] Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 30, mark them as medium-term bad users, and obtain the third training subset;

[0026] Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 1 and less than 15, mark them as medium-term gray users, and obtain the fourth training subset;

[0027] Among the long-term bad users, find the training samples that have not been overdue in the 1st to 3rd business periods, mark them as medium-term good users, and obtain the fifth training subset;

[0028] The long-term good users are taken as the sixth training subset.

[0029] Optionally, in a third implementation of the first aspect of the present invention, before obtaining a plurality of weight value samples and model hyperparameter samples, the method further includes:

[0030] Determine an initial weight value range and an initial model hyperparameter range according to the number of samples in each of the training subsets, select values ​​within the initial weight value range and the initial model hyperparameter range, and obtain multiple weight value samples and model hyperparameter samples;

[0031] The calculating the discrimination of the initial prediction model at each weight value sample and the discrimination of the model hyperparameter sample, and determining the weight search interval and the model hyperparameter search interval based on the discrimination comprises:

[0032] The samples of the test set are input into the initial prediction model and the discrimination at each weight value sample and each model hyperparameter sample under the short-term overdue label, the medium-term overdue label and the long-term overdue label is calculated, and the weight search interval and the model hyperparameter search interval are determined according to the changing trend of the discrimination.

[0033] Optionally, in a fourth implementation of the first aspect of the present invention, constructing an optimization objective function based on the weight search interval, the model hyperparameter search interval, and each of the discriminations corresponding to the test set and performing optimization calculations to obtain the optimal weight parameters and the optimal model hyperparameters for each of the training subsets includes:

[0034] An optimization objective function is constructed based on the weight search interval, the model hyperparameter search interval, and each of the discriminations corresponding to the test set. The expression of the optimization objective function is:

[0035] Target = A 1 *TestKS MOB6 +A 2 *TestKS MOB3 +A 3 *TestKS FPD15 ;

[0036] Among them, A 1 , A 2 and A 3 Focus on parameters for discrimination; TestKS MOB6 TestKS represents the discrimination of the model under long-term overdue labels in the test set. MOB3 TestKS represents the discrimination of the model under the mid-term overdue label in the test set. FPD15 represents the discrimination of the model under the short-term overdue labels in the test set;

[0037] Calling the Bayesian optimization algorithm to maximize the optimization objective function;

[0038] Obtain the weight parameters and model hyperparameters corresponding to when the function value of the optimization objective function is maximum, and obtain the optimal weight parameters and the optimal model hyperparameters.

[0039] Optionally, in a fifth implementation of the first aspect of the present invention, after inputting the data to be predicted into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted, the method further includes:

[0040] Regularly obtain the actual repayment data corresponding to the business forecast risk and the data to be forecasted;

[0041] Calculate the effectiveness index based on the business forecast risk and actual repayment data, and determine whether the effectiveness index meets the update requirements;

[0042] If the update requirements are met, the actual repayment data is captured, and the training samples are updated based on the actual repayment data and the risk prediction model is updated.

[0043] A second aspect of the present invention provides a business risk prediction device, comprising:

[0044] A sample clustering module is used to obtain multiple training samples and divide the training samples into a training set and a test set, and based on the numerical characteristics of each training sample, the training set is clustered according to the sample clustering rules to obtain multiple training subsets;

[0045] The model building module assigns weight parameters to the loss functions of multiple training subsets and calls the distributed gradient boosting algorithm to build the initial prediction model;

[0046] A weight calculation module is used to obtain multiple weight value samples and model hyperparameter samples, calculate the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample, and determine the weight search interval and the model hyperparameter search interval based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of short-term, medium-term and long-term; based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, an optimization objective function is constructed and an optimization calculation is performed to obtain the optimal weight parameters and the optimal model hyperparameters of each training subset;

[0047] A parameter adjustment module, used to adjust the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters, and obtain a trained risk prediction model after fitting the training set;

[0048] The risk prediction module is used to receive and respond to business risk prediction requests and obtain the data to be predicted; input the data to be predicted into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted.

[0049] The third aspect of the present invention provides a business risk prediction device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the business risk prediction device executes the steps of the above-mentioned business risk prediction method.

[0050] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned business risk prediction method.

[0051] A fifth aspect of the present invention provides a computer program product, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the business risk prediction method as described above are implemented.

[0052] In the technical solution provided by the present invention, a plurality of training samples are obtained, and the training samples are divided into a training set and a test set. Based on the numerical characteristics of each training sample, the training set is grouped according to the sample grouping rules to obtain a plurality of training subsets; weight parameters are assigned to the loss functions of the plurality of training subsets, and a distributed gradient boosting algorithm is called to build an initial prediction model; a plurality of weight value samples and model hyperparameter samples are obtained, the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample are calculated, and the weight search interval and the model hyperparameter search interval are determined based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of the short-term, medium-term and long-term; an optimization objective function is constructed based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, and an optimization calculation is performed to obtain the optimal weight parameters and the optimal model hyperparameters of each training subset; the initial prediction model is parameterized based on the optimal weight parameters and the optimal model hyperparameters, and a trained risk prediction model is obtained after fitting the training set; a business risk prediction request is received and responded to, and the data to be predicted is obtained; the data to be predicted is input into the trained risk prediction model for risk prediction, and the business prediction risk corresponding to the data to be predicted is obtained. The method can achieve good business risk prediction results in short-term, medium-term and long-term situations, and has strong cross-scenario capabilities; the model deployment complexity is low, and resource consumption is low. A business risk prediction device, electronic device, computer-readable storage medium and computer program product provided by the present invention also solve corresponding technical problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0054] Figure 1 A schematic diagram of the steps of an embodiment of a business risk prediction method in an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a processing flow of an embodiment of a business risk prediction method in an embodiment of the present invention;

[0056] Figure 3 A schematic diagram of a grouping rule in an embodiment of a business risk prediction method in an embodiment of the present invention;

[0057] Figure 4 A KS performance curve diagram of an embodiment of the business risk prediction method in an embodiment of the present invention;

[0058] Figure 5 A schematic diagram of an embodiment of a business risk prediction device in an embodiment of the present invention;

[0059] Figure 6 A schematic diagram of an embodiment of a business risk prediction device in an embodiment of the present invention;

[0060] Figure 7 The figure is a schematic diagram of a computer-readable medium in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] Exemplary embodiments of the present invention will now be described more fully with reference to the accompanying drawings. However, exemplary embodiments can be implemented in a variety of forms, and should not be construed as limiting the present invention to the embodiments set forth herein. On the contrary, providing these exemplary embodiments enables the present invention to be more comprehensive and complete, and is more convenient for fully conveying the inventive concept to those skilled in the art. The same reference numerals in the figures represent the same or similar elements, components or parts, and thus their repeated description will be omitted.

[0062] Under the premise of being consistent with the technical concept of the present invention, the features, structures, characteristics or other details described in a specific embodiment do not exclude that they can be combined in one or more other embodiments in a suitable manner.

[0063] In the description of specific embodiments, the features, structures, characteristics or other details described in the present invention are intended to enable those skilled in the art to fully understand the embodiments. However, it does not exclude that those skilled in the art can practice the technical solutions of the present invention without one or more of the specific features, structures, characteristics or other details.

[0064] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0065] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0066] The term "and / or" or "and / or" includes all combinations of any one or more of the associated listed items.

[0067] See also Figure 1-Figure 4 A specific embodiment of the business risk prediction method in the embodiment of the present invention includes:

[0068] S101, obtaining a plurality of training samples, and dividing the training samples into a training set and a test set, and clustering the training set according to a sample clustering rule based on the numerical characteristics of each training sample to obtain a plurality of training subsets;

[0069] It is understandable that the execution subject of the present invention may be a business risk prediction device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0070] In this embodiment, before the server receives a business risk prediction request, it is necessary to optimize and train the prediction model to be used in the prediction. In the field of business risk control, even if the bad samples obtained for training are all labeled as bad sample labels, they actually have certain characteristic differences; taking the prediction of repayment overdue risk as an example, even if the entire customer group corresponding to some training samples are all overdue users with long account ages, it does not mean that all users in this customer group have the same overdue risk at different account ages; the early and late overdue represent different overdue paths and different customer group portraits, so these customer groups should be treated differently, and the impact on the model objective function should also be treated differently, so as to improve the effect of predicting the user's overdue situation and obtain more accurate or effective business risk prediction results.

[0071] Specifically, taking the prediction of repayment overdue risk as an example, firstly, a plurality of training samples are obtained, wherein the training samples have different numerical characteristics, and the numerical characteristics of the training samples include business periods and business overdue days; in a specific implementation manner, the business periods can be in months.

[0072] First, the training samples are divided into a training set and a test set, and the samples in the training set are grouped according to the preset sample grouping rules to obtain multiple training subsets. The preset sample grouping rules can be to divide the training samples into groups according to the overdue situation to obtain training subsets marked with different overdue situation labels, specifically divided into training subsets with long-term overdue labels, training subsets with medium-term overdue labels, and training subsets with short-term overdue labels. In a specific implementation, the first period is short-term, the first three periods are medium-term, and the first six periods are long-term; overdue in the first period is short-term overdue, overdue in the second or third period is medium-term overdue, and overdue in the 4th to 6th period is long-term overdue.

[0073] Please continue reading Figure 3 , the specific grouping rules can be described as follows:

[0074] Step (1): Determine whether the training sample has any business overdue situation in the first 6 business periods. If there is no business overdue situation in the first 6 business periods, it is marked as a long-term good user and represented by "MOB6Ever30+good user"; if there is at least one business overdue situation in the first 6 business periods, and the number of business overdue days in any business overdue situation is greater than or equal to 30, it is marked as a long-term bad user and represented by "MOB6 Ever30+bad user";

[0075] Step (2): among the long-term bad users, find the training samples that are overdue in the first business period and whose business overdue days are greater than or equal to 15, mark them as short-term bad users, and obtain the first training subset, namely "FPD Ever15+ bad users";

[0076] Step (3): among the long-term bad users, find the training samples that are overdue in the first service period and whose service overdue days are greater than or equal to 1 and less than 15, mark them as short-term gray users, and obtain the second training subset, namely "FPD Ever15+ gray users";

[0077] Step (4): among the long-term bad users, find the training samples that have no overdue payment in the first business period, have overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 30, and obtain the third training subset, which is marked as medium-term bad users, i.e., "MOB3 Ever30+ bad users";

[0078] Step (5): among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 1 and less than 15, and obtain the fourth training subset, which is marked as medium-term gray users, namely "MOB3 Ever30+ gray users";

[0079] Step (6): among the long-term bad users, find the training samples that have not been overdue in the 1st to 3rd service periods, and obtain the fifth training subset, which is marked as medium-term good users, i.e., "MOB3 Ever30+ good users";

[0080] Step (7): The long-term good users are used as the sixth training subset, namely, "MOB6 Ever30+ good users".

[0081] The training samples in the training set are divided into multiple training subsets through the above-mentioned clustering rules, and each training subset represents a different customer group, and a different weight is assigned to each customer group, where the weight assigned in this step is the initial weight value; and the label of a long-term good user is marked as 0, and the label of a long-term bad user is marked as 1, so as to be used for distinction in subsequent training.

[0082] S102, assigning weight parameters to the loss functions of multiple training subsets, and calling the distributed gradient boosting algorithm to build an initial prediction model;

[0083] Among them, the distributed gradient boosting algorithm used in this embodiment can be the XGBoost (Extreme Gradient Boosting) algorithm, which is an optimized distributed gradient boosting library designed to achieve high efficiency, flexibility and portability; it can implement machine learning algorithms under the Gradient Boosting framework, provides parallel tree boosting, and can quickly and accurately solve many data science problems.

[0084] Specifically, the XGBoost algorithm is first called to generate an initial objective function. In a specific implementation, an initial objective function can be expressed as:

[0085]

[0086] Where obj(θ) represents the initial objective function of the prediction model; θ represents the model parameters; n represents the number of training samples; LOSS is the loss function; y i Represents the true value of the i-th training sample; Indicates that the model is based on the input x i The predicted value of the i-th training sample is ω(f k ) represents the complexity of the kth subtree, and K is the total number of subtrees.

[0087] Subsequently, the initial objective function is revised based on the initial weight value assigned in the previous step to generate a weighted objective function, the expression of which is:

[0088]

[0089] Here, obj(θ) represents the weighted objective function of the prediction model, θ represents the model parameters, m represents the total number of training subsets, and ω represents the weighted objective function of the prediction model. j is the weight parameter of the jth training subset; n j is the number of samples in the jth training subset, where j = 1, 2, 3, …, m; y i Represents the true value of the i-th training sample; represents the value of the i-th training sample predicted by the model based on the input; ω(f k ) represents the complexity of the kth subtree, and K is the total number of subtrees.

[0090] In the specific implementation as described in step S101, when the number of training subsets is 6, m=6 in the weighted objective function.

[0091] S103, obtaining a plurality of weight value samples and model hyperparameter samples, and calculating the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample, and determining the weight search interval and the model hyperparameter search interval based on the discrimination;

[0092] In this embodiment, a more precise search range is determined by obtaining a weight search interval and a model hyperparameter search interval, so as to facilitate obtaining the optimal weight parameters and the optimal model hyperparameters in subsequent steps; when determining the weight search interval and the model hyperparameter search interval, firstly, multiple weight value samples and model hyperparameter samples are selected, and based on these weight value samples and model hyperparameter samples, the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample are calculated, and the weight search interval and model hyperparameter search interval where the optimal weight value and the optimal model hyperparameter may be located are obtained according to the discrimination.

[0093] In this step, obtaining the weight search interval is taken as an example. The weight of long-term good users is set to 1, and other weights need to be set separately according to the customer groups in the training subset obtained above. Specifically, after the above steps have completed the construction of the prediction model and obtained the initial weight values, the actual weight parameters of the weighted objective function in the model can be determined based on each training sample, where the actual weight parameters can be selected by calculating the discrimination of the initial prediction model at each weight value, and the weight value with better discrimination effect is taken as the final actual weight parameter; and, when the training set is updated, the actual weight parameter can be re-acquired based on the updated training set.

[0094] Specifically, first determine the initial weight value range according to the number of samples in each of the training subsets, take values ​​within the initial weight value range, and obtain multiple weight value samples. Obtain the training set generated in the above steps, and based on a classification method similar to that in step S101, label each training sample contained in the training set to obtain a test set of training samples labeled with short-term overdue labels, medium-term overdue labels, and long-term overdue labels. In the initial prediction model constructed based on the input of each weight value sample, calculate the discrimination of the initial prediction model at each weight value sample, wherein the discrimination includes the discrimination of the model of the test set under the short-term, medium-term, and long-term overdue labels. According to the changing trend of the discrimination, the optimal weight search interval under different overdue labels can be determined, wherein the discrimination can be measured by the KS (Kolmogorov–Smirnov) test performance.

[0095] In a specific implementation, first, the initial weight value range of each training subset is determined by the data distribution characteristics in the training samples. Specifically, the minimum value can be set to 1, and the maximum value can be set to the ratio of the number of long-term good samples to the number of samples in the current training subset. For example, if the number of good samples in the training set is 10,000 and the number of samples of customer group 1 is 1,000, then ω 1 The initial weight value range is [1,10], that is, the minimum value is 1 and the maximum value is 10. Then, within the initial weight value range, an arithmetic sequence is generated at fixed intervals, and the arithmetic sequence is used as the sample value of the initial weight value, which is used as ω in turn. 1 Input the model and calculate the KS performance values ​​of the test set under the short-term, medium-term and long-term overdue labels, that is, the KS performance values ​​of the test set under different prediction values ​​of "FPDEver15+", "MOB3 Ever30+" and "MOB6 Ever30+", and draw the KS performance curve. A specific KS performance curve can be shown as follows Figure 4 As shown. Find the weight search interval according to the change of KS performance value in the KS performance curve. For a specific example, please continue to refer to Figure 4 , the training samples are divided into training set and test set and the KS performance value under each overdue label is calculated. Taking the long-term overdue label as an example, when ω 1 As ω increases gradually, the long-term overdue KS performance value of the training set gradually decreases, but the long-term overdue KS performance value of the test set remains stable in the initial stage. 1 When it is greater than 5.5, it decays significantly; while the FPD15 KS performance value gradually increases and gradually stabilizes when it is greater than 7; while the MOB3KS remains basically stable; therefore, 1The search interval can be further narrowed to [1,5.5]. In this interval, the long-term KS performance value of the test set has basically no decay, while the short-term overdue KS performance value increases significantly.

[0096] Based on the above-mentioned weight search method, the method for obtaining the model hyperparameter search interval is also similar. First, determine the initial model hyperparameter range, take values ​​within the initial model hyperparameter range, and obtain multiple model hyperparameter samples. Obtain the training set generated in the previous steps, and based on a classification method similar to that in step S101, label each training sample contained in the training set to obtain a test set of training samples labeled with short-term overdue labels, medium-term overdue labels, and long-term overdue labels. In the initial prediction model constructed based on the input of each model hyperparameter sample, calculate the discrimination of the initial prediction model at each model hyperparameter sample, wherein the discrimination includes the discrimination of the model of the test set under the short-term, medium-term, and long-term overdue labels. According to the changing trend of the discrimination, the optimal model hyperparameter search interval under different overdue labels can be determined. In addition, other weights of the customer groups corresponding to the training subsets can also be used in a similar manner to determine a more refined search range.

[0097] In a specific embodiment, the model hyperparameters may be XGBoost hyperparameters.

[0098] S104, constructing an optimization objective function based on the weight search interval, the model hyperparameter search interval, and the discrimination corresponding to the test set, and performing optimization calculation to obtain the optimal weight parameters and the optimal model hyperparameters for each training subset;

[0099] The corresponding discrimination described in this step refers to the discrimination data of the initial prediction model for the samples in the test set under the respective overdue labels of short-term, medium-term and long-term within the obtained weight search interval and model hyperparameter search interval. After obtaining the weight search interval and model hyperparameter search interval, the discrimination data of the samples contained in the test set within the weight search interval and model hyperparameter search interval are calculated, and an optimization objective function is constructed based on each discrimination. The expression of the optimization objective function can be:

[0100] Target = A 1 *TestKS MOB6 +A 2 *TestKS MOB3 +A 3 *TestKS FPD15 ;

[0101] Among them, A 1 , A 2 and A 3 For discrimination, TestKS focuses on parameters. MOB6Indicates the long-term discrimination of the model on the test set, which is used to measure the long-term performance of the model. TestKS MOB3 Indicates the mid-term discrimination of the model on the test set, which is used to measure the performance of the model in the mid-term, TestKS FPD15 Indicates the short-term discrimination of the model on the test set, which is used to measure the performance of the model in the short term. 1 , A 2 and A 3 represents the business emphasis on different degrees of differentiation under different deadlines, and the cumulative sum is equal to 1; it can be a constant set according to statistical methods, based on business needs or based on experience. In a specific example, A 1 =0.5, A 2 =0.3, A 3 =0.2; at this value, the optimization objective function focuses more on long-term discrimination.

[0102] After obtaining the optimization objective function, the Bayesian optimization algorithm is called to maximize the optimization objective function; wherein A 1 , A 2 and A 3 is a fixed value, TestKS MOB6 、TestKS MOB3 and TestKS FPD15 It is the value of the model under different weight parameters and model hyperparameters. When the Bayesian optimization algorithm is called to maximize the optimization objective function, the weight parameters and model hyperparameters corresponding to the maximum function value of the optimization objective function can be obtained, and the optimal weight parameters and the optimal model hyperparameters can be obtained. Among them, the hyperparameters include the maximum tree depth (max_depth), minimum segmentation benefit (gamma) and other hyperparameters that come with XGBoost itself.

[0103] S105, adjusting the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters to obtain a trained risk prediction model;

[0104] Based on the optimal weight parameters and optimal model hyperparameters obtained in the previous steps, the model parameters are adjusted and the training set is fitted to obtain the trained risk prediction model.

[0105] Among them, the simplified function expression of the initial prediction model is: In the "model", the specific output is determined by the initial prediction model parameter θ, which controls how the model generates prediction values ​​from input features. Therefore, after obtaining the specific weighting parameters of the weighted objective function and the model hyperparameters, the parameters of the initial prediction model are adjusted and the training set is fitted so that the adjusted parameters can take effect and a trained risk prediction model is obtained.

[0106] S106, receiving and responding to a business risk prediction request, and obtaining data to be predicted;

[0107] S107: Input the data to be predicted into the trained risk prediction model to perform risk prediction, and obtain the business prediction risk corresponding to the data to be predicted.

[0108] After obtaining the risk prediction model, it is possible to receive and respond to business risk prediction requests to obtain data to be predicted, wherein the data to be predicted can be single data or batch data; after the data to be predicted is input into the trained risk prediction model, the model will perform risk prediction based on the characteristics of the data to be predicted to obtain the relative size of the business prediction risk corresponding to the data to be predicted or the ranking information of the risk score; in a specific embodiment, the user's overdue risk level or risk ranking information can be inferred based on the characteristics of the data to be predicted, and this overdue risk can not only focus on long-term overdue risks, but also medium-term and short-term overdue risks.

[0109] In addition, the business risk prediction method in this embodiment can also regularly obtain the actual repayment data corresponding to the business prediction risk and the data to be predicted; calculate the effectiveness index of the model according to the business prediction risk and the actual repayment data, and the effectiveness index can include discrimination indexes such as KS and other indicators, and can also include other indicators, such as AUC (Area UnderCurve, area under the receiver operating characteristic curve); and judge whether the effectiveness index meets the update requirements; if the update requirements are met, the actual repayment data is captured, and the training samples are updated based on the actual repayment data and the risk prediction model is updated; wherein, the update requirement can be that the KS or AUC of the model on the latest sample has obvious attenuation. And, when the time from the last update meets the update duration, the actual repayment data is captured to update the training samples and update the risk prediction model.

[0110] The business risk prediction method in the embodiment of the present invention can output more accurate business risk prediction results with better prediction effectiveness in short-term, medium-term and long-term situations; and the model deployment has low complexity and low resource consumption. Further, the method can significantly improve the prediction performance in the medium-term and short-term performance period while maintaining the long-term prediction performance by giving higher weights to medium-term and short-term overdue behaviors, enhance the timeliness and early warning ability of the model, enable financial institutions to identify and respond to potential overdue risks earlier, and help reduce the bad debt rate; the model can more accurately capture the typical characteristics of different overdue behavior patterns, thereby improving the generalization performance and recognition accuracy of the model in various overdue scenarios, and has strong cross-scenario capabilities, achieving accurate identification and effective control of risks, and providing strong support for the risk decision-making of financial institutions; and by setting a higher weight for bad samples, the model focuses more on learning and identifying bad sample features during the training process, thereby showing stronger bad sample recognition ability and risk interception effect in practical applications; compared with the traditional method that requires training and maintaining multiple models to deal with different overdue stages, the present invention can achieve effective prediction of short-term and long-term overdue through only one model, greatly reducing resource consumption and the complexity of model deployment. This advantage not only reduces the cost of model development and maintenance, but also simplifies the risk management process and improves the operational efficiency of financial institutions.

[0111] The above describes the business risk prediction method in the embodiment of the present invention. The following describes the business risk prediction device in the embodiment of the present invention. Figure 5 , an embodiment of the business risk prediction device in the embodiment of the present invention includes:

[0112] The sample clustering module 501 is used to obtain multiple training samples, and divide the training samples into a training set and a test set, and cluster the training set according to the sample clustering rules based on the numerical characteristics of each training sample to obtain multiple training subsets;

[0113] A model building module 502 is used to assign weight parameters to the loss functions of multiple training subsets according to the number of subsets of the training subsets, and call a distributed gradient boosting algorithm to build an initial prediction model;

[0114] The weight calculation module 503 is used to obtain multiple weight value samples and model hyperparameter samples, and calculate the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample, and determine the weight search interval and the model hyperparameter search interval based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of short-term, medium-term and long-term; construct an initial optimization objective function based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, and perform optimization calculation to obtain the optimal weight parameters and the optimal model hyperparameters of each training subset;

[0115] A parameter adjustment module 504 is used to adjust the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters, and obtain a trained risk prediction model after fitting the training set;

[0116] The risk prediction module 505 is used to receive and respond to business risk prediction requests, obtain the data to be predicted, input the data to be predicted into the trained risk prediction model for risk prediction, and obtain the business prediction risk corresponding to the data to be predicted.

[0117] The business risk prediction device in the embodiment of the present invention has good business risk prediction effects in short-term, medium-term and long-term situations; and the complexity of model deployment is low and resource consumption is low.

[0118] In another embodiment of the present application, the distributed gradient boosting algorithm is an XGBoost algorithm, and the weighted objective function expression included in the initial prediction model is:

[0119]

[0120] Where obj(θ) represents the weighted objective function of the prediction model; θ represents the model parameters; m represents the total number of training subsets; ω j is the weight parameter of the jth training subset; n j is the number of samples in the jth training subset, where j = 1, 2, 3, …, m; y i Represents the true value of the i-th training sample; represents the value of the i-th training sample predicted by the model based on the input; ω(f k ) represents the complexity of the kth subtree, and K is the total number of subtrees.

[0121] In another embodiment of the present application, the numerical features of the training samples include the number of business periods and the number of business overdue days;

[0122] The sample grouping module 501 is specifically used for:

[0123] The training samples in the training set are grouped and labeled according to the business period and overdue days. If there are no business overdue days in the first 6 business periods, they are labeled as long-term good users;

[0124] If the number of overdue days in the first 6 business periods is greater than or equal to 30, the user will be marked as a long-term bad user;

[0125] Among the long-term bad users, find the training samples that are overdue in the first business period and the number of business overdue days is greater than or equal to 15, mark them as short-term bad users, and obtain the first training subset;

[0126] Among the long-term bad users, find the training samples that are overdue in the first business period and whose business overdue days are greater than or equal to 1 and less than 15, mark them as short-term gray users, and obtain the second training subset;

[0127] Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 30, mark them as medium-term bad users, and obtain the third training subset;

[0128] Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 1 and less than 15, mark them as medium-term gray users, and obtain the fourth training subset;

[0129] Among the long-term bad users, find the training samples that have not been overdue in the 1st to 3rd business periods, mark them as medium-term good users, and obtain the fifth training subset;

[0130] The long-term good users are taken as the sixth training subset.

[0131] In another embodiment of the present application, the weight calculation module 503, before obtaining the plurality of weight value samples, further includes:

[0132] Determine an initial weight value range and an initial model hyperparameter range according to the number of samples in each of the training subsets, select values ​​within the initial weight value range and the initial model hyperparameter range, and obtain multiple weight value samples and model hyperparameter samples;

[0133] The weight calculation module 503 is also used to input the samples of the test set into the initial prediction model and calculate the discrimination at each weight value sample and the discrimination at the model hyperparameter sample under the short-term overdue label, the medium-term overdue label and the long-term overdue label, and determine the weight search interval and the model hyperparameter search interval according to the changing trend of the discrimination.

[0134] In another embodiment of the present application, the weight calculation module 503 is further configured to:

[0135] An optimization objective function is constructed based on the weight search interval, the model hyperparameter search interval, and the discrimination corresponding to the test set. The expression of the optimization objective function is:

[0136] Target = A 1 *TestKS MOB6 +A 2 *TestKS MOB3 +A 3 *TestKS FPD15 ;

[0137] Among them, A 1 , A 2 and A 3 Focus on parameters for discrimination; TestKS MOB6 TestKS represents the discrimination of the model under long-term overdue labels in the test set. MOB3 TestKS represents the discrimination of the model under the mid-term overdue label in the test set. FPD15 represents the discrimination of the model under the short-term overdue labels in the test set;

[0138] Calling the Bayesian optimization algorithm to maximize the optimization objective function;

[0139] Obtain the weight parameters and model hyperparameters corresponding to when the function value of the optimization objective function is maximum, and obtain the optimal weight parameters and the optimal model hyperparameters.

[0140] In another embodiment of the present application, the business risk prediction device further includes a model updating module, and the model updating module is specifically used to:

[0141] Regularly obtain the actual repayment data corresponding to the business forecast risk and the data to be forecasted;

[0142] Calculate the effectiveness index based on the business forecast risk and actual repayment data, and determine whether the effectiveness index meets the update requirements;

[0143] If the update requirements are met, the actual repayment data is captured, and the training samples are updated based on the actual repayment data and the risk prediction model is updated.

[0144] The business risk prediction device in the embodiment of the present invention can achieve good business risk prediction results in a variety of short-term, medium-term and long-term situations, and has strong cross-scenario capabilities; the model deployment complexity is low, and resource consumption is low.

[0145] The present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the business risk prediction method described in any of the above embodiments.

[0146] Based on the same inventive concept, the embodiment of this specification also provides a business risk prediction device. The business risk prediction device in the embodiment of the present invention is described in detail below from the perspective of hardware processing.

[0147] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this specification. Figure 6 The electronic device 600 according to this embodiment of the present invention is described. Figure 6 The electronic device 600 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0148] like Figure 6 As shown, the electronic device 600 is in the form of a general computing device. The components of the electronic device 600 may include, but are not limited to: at least one processing unit 610, at least one storage unit 620, a bus 630 connecting different system components (including the storage unit 620 and the processing unit 610), a display unit 640, etc.

[0149] The storage unit stores program codes, which can be executed by the processing unit 610, so that the processing unit 610 performs the steps according to various exemplary embodiments of the present invention described in the above processing method section of this specification. For example, the processing unit 610 can perform the following steps: Figure 1 or Figure 2 Steps shown.

[0150] The storage unit 620 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 6201 and / or a cache memory unit 6202 , and may further include a read-only memory unit (ROM) 6203 .

[0151] The storage unit 620 may also include a program / utility 6204 having a set (at least one) of program modules 6205, such program modules 6205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each of which or some combination may include the implementation of a network environment.

[0152] Bus 630 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0153] The electronic device 600 may also communicate with one or more external devices 100 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device 600, and / or any device that enables the electronic device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed through an input / output (I / O) interface 650. Furthermore, the electronic device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 660. The network adapter 660 may communicate with other modules of the electronic device 600 through the bus 630. It should be understood that although Figure 6 Not shown, other hardware and / or software modules may be used in conjunction with electronic device 600, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0154] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the exemplary embodiments described in the present invention can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation method of the present invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.) or on a network, including a number of instructions to enable a computing device (which can be a personal computer, server, or network device, etc.) to execute the above method according to the present invention. When the computer program is executed by a data processing device, the computer-readable medium can implement the above method of the present invention, that is: Figure 1 or Figure 2 The method shown.

[0155] Figure 7 A schematic diagram of a computer-readable medium provided for an embodiment of this specification.

[0156] accomplish Figure 1 or Figure 2The computer program of the method shown can be stored on one or more computer readable media. The computer readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0157] The computer readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by an instruction execution system, an apparatus, or a device or used in combination with it. The program code contained on the readable storage medium may be transmitted with any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0158] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0159] In summary, the present invention can be implemented in hardware, or in a software module running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that general data processing devices such as microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and a computer program product) for executing part or all of the methods described herein. Such a program implementing the present invention can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0160] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the present invention is not inherently related to any specific computer, virtual device or electronic device, and various general devices can also implement the present invention. The above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0161] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0162] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A business risk prediction method, characterized in that: include: Acquire multiple training samples, and divide the training samples into a training set and a test set, and based on the numerical characteristics of each of the training samples, perform sample grouping on the training set according to a sample grouping rule to obtain multiple training subsets; Assign weight parameters to the loss functions of multiple training subsets and call the distributed gradient boosting algorithm to build the initial prediction model; Obtain multiple weight value samples and model hyperparameter samples, calculate the discrimination of the initial prediction model at each weight value sample and the discrimination of the model hyperparameter sample, and determine the weight search interval and the model hyperparameter search interval based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of short-term, medium-term and long-term; Based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, an optimization objective function is constructed and an optimization calculation is performed to obtain the optimal weight parameters and the optimal model hyperparameters for each of the training subsets; Adjusting the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters, and obtaining a trained risk prediction model after fitting the training set; Receive and respond to business risk prediction requests and obtain data to be predicted; The data to be predicted is input into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted.

2. The business risk prediction method according to claim 1, characterized in that: The distributed gradient boosting algorithm is the XGBoost algorithm, and the weighted objective function expression contained in the initial prediction model is: Where obj(θ) represents the weighted objective function of the prediction model; θ represents the model parameters; m represents the total number of training subsets; ω j is the weight parameter of the jth training subset; n j is the number of samples in the jth training subset, where j = 1, 2, 3, …, m; y i Represents the true value of the i-th training sample; represents the value of the i-th training sample predicted by the model based on the input; ω(f k ) represents the complexity of the kth subtree, and K is the total number of subtrees.

3. The business risk prediction method according to claim 2, characterized in that: The numerical features of the training samples include the number of business periods and the number of business overdue days; Based on the numerical features of each training sample, the training set is grouped according to the sample grouping rule to obtain multiple training subsets including: The training samples in the training set are grouped and labeled according to the business period and the number of overdue days. If there are no business overdue days in the first 6 business periods, they are labeled as long-term good users; if the business overdue days in the first 6 business periods are greater than or equal to 30, they are labeled as long-term bad users; Among the long-term bad users, find the training samples that are overdue in the first business period and the number of business overdue days is greater than or equal to 15, mark them as short-term bad users, and obtain the first training subset; Among the long-term bad users, find the training samples that are overdue in the first business period and whose business overdue days are greater than or equal to 1 and less than 15, mark them as short-term gray users, and obtain the second training subset; Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 30, mark them as medium-term bad users, and obtain the third training subset; Among the long-term bad users, find the training samples that have no overdue payment in the first business period, overdue payment in the second business period or the third business period, and the number of overdue days is greater than or equal to 1 and less than 15, mark them as medium-term gray users, and obtain the fourth training subset; Among the long-term bad users, find the training samples that have not been overdue in the 1st to 3rd business periods, mark them as medium-term good users, and obtain the fifth training subset; The long-term good users are taken as the sixth training subset.

4. The business risk prediction method according to claim 3, characterized in that: Before obtaining a plurality of weight value samples and model hyperparameter samples, the method further includes: Determine an initial weight value range and an initial model hyperparameter range according to the number of samples in each of the training subsets, select values ​​within the initial weight value range and the initial model hyperparameter range, and obtain multiple weight value samples and model hyperparameter samples; The calculating the discrimination of the initial prediction model at each weight value sample and the discrimination of the model hyperparameter sample, and determining the weight search interval and the model hyperparameter search interval based on the discrimination comprises: The samples of the test set are input into the initial prediction model and the discrimination at each weight value sample and each model hyperparameter sample under the labels of short-term overdue, medium-term overdue and long-term overdue is calculated, and the weight search interval and model hyperparameter search interval are determined according to the changing trend of the discrimination.

5. The business risk prediction method according to claim 4, characterized in that: The optimization objective function is constructed based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, and the optimization calculation is performed to obtain the best weight parameter and the best model hyperparameter of each training subset, which includes: An optimization objective function is constructed based on the weight search interval, the model hyperparameter search interval, and each of the discriminations corresponding to the test set. The expression of the optimization objective function is: <h2 style=";text-align:left;direction:ltr">Target=A1*TestKS<h2 style=";text-align:left;direction:ltr"> MOB6 <h2 style=";text-align:left;direction:ltr"> +A2*TestKS<h2 style=";text-align:left;direction:ltr"> MOB3 <h2 style=";text-align:left;direction:ltr"> +A3*TestKS<h2 style=";text-align:left;direction:ltr"> FPD15 <h2 style=";text-align:left;direction:ltr"> ; Among them, A1, A2 and A3 are discrimination emphasis parameters; among them, TestKS MOB6 Indicates the long-term discrimination of the model on the test set, TestKS MOB3 Indicates the mid-term discrimination of the model on the test set, TestKS FPD15 Indicates the short-term discrimination of the model on the test set; Calling the Bayesian optimization algorithm to maximize the optimization objective function; Obtain the weight parameters and model hyperparameters corresponding to when the function value of the optimization objective function is maximum, and obtain the optimal weight parameters and the optimal model hyperparameters.

6. The business risk prediction method according to claim 5, characterized in that: After inputting the data to be predicted into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted, the method further includes: Regularly obtain the actual repayment data corresponding to the business forecast risk and the data to be forecasted; Calculate the effectiveness index based on the business forecast risk and actual repayment data, and determine whether the effectiveness index meets the update requirements; If the update requirements are met, the actual repayment data is captured, the training samples are updated based on the actual repayment data, and the risk prediction model is updated.

7. A business risk prediction device, characterized in that: The business risk prediction device comprises: A sample clustering module is used to obtain multiple training samples and divide the training samples into a training set and a test set, and based on the numerical characteristics of each training sample, the training set is clustered according to the sample clustering rules to obtain multiple training subsets; The model building module assigns weight parameters to the loss functions of multiple training subsets and calls the distributed gradient boosting algorithm to build the initial prediction model; A weight calculation module is used to obtain multiple weight value samples and model hyperparameter samples, calculate the discrimination of the initial prediction model at each weight value sample and the discrimination at the model hyperparameter sample, and determine the weight search interval and the model hyperparameter search interval based on the discrimination, wherein the discrimination is the discrimination of the test set under the respective overdue labels of short-term, medium-term and long-term; based on the weight search interval, the model hyperparameter search interval and the discrimination corresponding to the test set, an optimization objective function is constructed and an optimization calculation is performed to obtain the optimal weight parameters and the optimal model hyperparameters of each training subset; A parameter adjustment module, used to adjust the parameters of the initial prediction model based on the optimal weight parameters and the optimal model hyperparameters, and obtain a trained risk prediction model after fitting the training set; The risk prediction module is used to receive and respond to business risk prediction requests and obtain the data to be predicted; input the data to be predicted into the trained risk prediction model for risk prediction to obtain the business prediction risk corresponding to the data to be predicted.

8. A business risk prediction device, characterized in that: The business risk prediction device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the business risk prediction device executes the steps of the business risk prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the program / instructions are executed by the processor, the steps of the business risk prediction method as described in any one of claims 1-6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the business risk prediction method as described in any one of claims 1 to 6 are implemented.

Citation Information

Cited By

  • Overdue risk identification method and device and computer storage medium

    CN120146984A

  • Risk prediction and training method of dichotomy model based on sorting loss and related products

    CN120257108A