Loan default prediction method and device

Abstract

The invention provides a loan default prediction method and device, and relates to the technical field of risk assessment. The method comprises the steps that historical data are acquired, the historical data comprise multiple pieces of historical user loan data, and each piece of historical user loan data comprises basic features and a corresponding loan default label; clustering the historical data, and performing default risk assessment on each clustered data set to obtain a risk level of each clustered data set; determining an oversampling strategy of each clustering data set based on the risk level of each clustering data set; on the basis of the oversampling strategy of each clustering data set, oversampling is carried out on each clustering data set; and based on the balanced data set, training a preset default prediction model to obtain a loan default prediction model for loan default prediction. And blind user manufacturing in a high-risk group is avoided, so that the model is more stable, and the loan default prediction effect is improved.

Description

Technical Field

[0001] This invention relates to the field of risk assessment technology, and more specifically, to a method and apparatus for predicting loan defaults. Background Technology

[0002] Auto loan business is a crucial segment for financial institutions, and loan default prediction is a key factor in loan disbursement. Currently, the primary method for loan default prediction is machine learning-based, which trains models using historical data. However, the vast majority of users in this historical data repay on time, with only a small minority defaulting, leading to a severe class imbalance in the training data. Directly using this data to train machine learning models results in a heavily biased bias towards the majority class (reliable users), leading to extremely poor predictive performance for the minority class (defaulting users) and an inability to effectively identify potential risks.

[0003] Currently, the industry commonly uses oversampling techniques, such as the SMOTE algorithm, to address such problems. However, traditional oversampling techniques have significant drawbacks: they typically generate samples globally or randomly, and loan users can naturally form different groups based on their income level, loan conditions, vehicle attributes, etc., with these groups exhibiting drastically different default characteristics and risk drivers. Traditional global oversampling methods generate a large number of business-irrational "noise samples" among groups with different risk characteristics, blurring the true decision boundaries and even leading to a decline in model performance. Summary of the Invention

[0004] The purpose of this invention is to provide a loan default prediction method and apparatus to improve the problem in the prior art of generating a large number of "noise samples" that are not business-reasonable among groups with different risk characteristics, which blurs the real decision boundary and even leads to a decline in model performance.

[0005] The first aspect of this application provides a method for predicting loan defaults, including the following steps: Acquire historical data, which includes multiple historical user loan data. Each historical user loan data includes basic characteristics and a corresponding loan default tag. The basic characteristics include at least: user characteristics, loan characteristics, and vehicle transaction characteristics. The historical data is clustered to obtain multiple clustered datasets, each with different default risk characteristics; Each cluster dataset is assessed for default risk to obtain its risk level. Based on the risk level of each cluster dataset, the oversampling strategy for each cluster dataset is determined. Based on the oversampling strategy of each cluster dataset, oversampling is performed on each cluster dataset to obtain a balanced dataset; Based on the balanced dataset, a pre-set default prediction model is trained to obtain a loan default prediction model for use in loan default prediction.

[0006] In this embodiment of the application, the clustering of the historical data to obtain multiple clustered datasets includes: Based on the basic features in the loan data of each historical user, composite features for each historical user are constructed according to preset composite feature construction rules. Based on the historical data and the composite features of each historical user, an initial training dataset is obtained; The initial training dataset is clustered to obtain multiple clustered datasets.

[0007] In this embodiment of the application, the step of constructing composite features for each historical user based on the basic features in the loan data of each historical user according to preset composite feature construction rules includes: Based on the loan characteristics and user characteristics in each historical user loan data, a first composite feature and a second composite feature are constructed respectively. The first composite feature is used to assess the user's debt pressure and repayment ability, and the second composite feature is used to assess the user's historical credit. Based on the loan characteristics in the loan data of each historical user, a third composite feature is constructed, which is used to assess the frequency with which a user seeks a new loan. Based on the user characteristics in each of the historical user loan data, a fourth composite feature is constructed. The fourth composite feature is used to represent the completeness and stability of the user's application materials. Based on the first composite feature, the second composite feature, the third composite feature, and the fourth composite feature, the composite features of each historical user are obtained.

[0008] In this embodiment of the application, the clustering of the initial training dataset to obtain multiple clustered datasets includes: The initial training dataset was clustered using the K-Means clustering algorithm to obtain multiple clustered datasets.

[0009] In this embodiment of the application, the step of assessing the default risk of each clustered dataset to obtain the risk level of each clustered dataset includes: The historical default rate of each cluster dataset is calculated based on the loan default labels in each cluster dataset. The historical default rate of each cluster dataset is compared with a preset default threshold to obtain the risk level of each cluster dataset.

[0010] In this embodiment of the application, the step of calculating the historical default rate of each cluster dataset based on the loan default labels in each cluster dataset includes: The number of defaulting users in each cluster dataset is calculated based on the loan default labels in each cluster dataset. The historical default rate of each cluster dataset is obtained based on the ratio of the number of defaulting users to the total number of users in each cluster dataset.

[0011] In this embodiment of the application, the step of oversampling each clustered dataset based on the oversampling strategy of each clustered dataset to obtain a balanced dataset includes: Based on the oversampling strategy of each cluster dataset, the oversampling factor of each cluster dataset is obtained; Based on the oversampling factor of each cluster dataset, the SMOTE algorithm is used to oversample each cluster dataset to obtain a balanced dataset.

[0012] A second aspect of this application provides a loan default prediction device, comprising: The acquisition module is used to acquire historical data, which includes multiple historical user loan data. Each historical user loan data includes basic characteristics and a corresponding loan default tag. The basic characteristics include at least: user characteristics, loan characteristics, and vehicle transaction characteristics. The clustering module is used to cluster the historical data to obtain multiple clustered datasets, each with different default risk characteristics. The assessment module is used to assess the default risk of each clustered dataset separately and obtain the risk level of each clustered dataset. The determination module is used to determine the oversampling strategy for each cluster dataset based on the risk level of each cluster dataset; The oversampling module is used to oversample each cluster dataset according to the oversampling strategy of each cluster dataset to obtain a balanced dataset. The training module is used to train a pre-set default prediction model based on the balanced dataset to obtain a loan default prediction model for use in loan default prediction.

[0013] The embodiments of the present invention have at least the following advantages or beneficial effects: The above technical solution involves acquiring historical data, including multiple historical user loan data sets. Each historical user loan data set includes basic features and a corresponding loan default label. The basic features include at least user characteristics, loan characteristics, and vehicle transaction characteristics. The historical data is then clustered to obtain multiple cluster datasets, each with different default risk characteristics. Default risk assessments are performed on each cluster dataset to obtain its risk level. Based on the risk levels of each cluster dataset, an oversampling strategy is determined. Based on the oversampling strategy, each cluster dataset is oversampled to obtain a balanced dataset. A pre-set default prediction model is trained on the balanced dataset to obtain a loan default prediction model for loan default prediction. By introducing a risk adaptive mechanism, different oversampling intensities are applied to clusters with different risks, avoiding the blind creation of users in high-risk groups, making the model more robust and conforming to the principle of prudent risk control. Within the framework of clustering and oversampling, domain feature engineering and risk adaptive strategies are used to address the issue of poor performance in specific business scenarios, thereby improving the effectiveness of loan default prediction. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart of a loan default prediction method provided in an embodiment of the present invention; Figure 2 This is a flowchart of the auto loan default prediction steps provided in an embodiment of the present invention; Figure 3 A structural block diagram of a loan default prediction device provided in an embodiment of the present invention; Figure 4 This is a structural block diagram of an electronic device provided in an embodiment of the present invention.

[0016] Icons: 410 - Acquisition module; 420 - Clustering module; 430 - Evaluation module; 440 - Determination module; 450 - Oversampling module; 460 - Training module; 101 - Memory; 102 - Processor; 103 - Communication interface. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0018] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely to distinguish one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0020] In the description of this application, it should be noted that the terms "upper", "lower", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship that the product of this application is usually placed in. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0021] The acquisition, transmission, storage, use, and processing of data in this application comply with relevant national laws and regulations. Furthermore, it should be noted that existing industry solutions such as software, components, and models may be mentioned in the embodiments of this application. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0022] Please refer to Figure 1 and Figure 2 , Figure 1This is a flowchart of a loan default prediction method provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating the steps for predicting auto loan defaults according to an embodiment of the present invention. This embodiment provides a method for predicting loan defaults, including the following steps: Step S210: Obtain historical data, which includes multiple historical user loan data. Each historical user loan data includes basic characteristics and a corresponding loan default tag. The basic characteristics include at least: user characteristics, loan characteristics, and vehicle transaction characteristics. In this embodiment, multiple historical user loan data sets can be used directly as historical data, or preprocessed data from multiple historical user loan data sets can be used as historical data. The preprocessing includes, but is not limited to, cleaning, missing value imputation, and standardization. Multiple historical user loan data sets can be obtained by acquiring historical user loan data from multiple users. The loan characteristics mentioned above can include the loan amount, asset cost, loan-to-asset ratio, as well as the user's historical total monthly payments, outstanding loan balance, number of overdue payments, and recent default records, which can be used to comprehensively assess the user's debt pressure and repayment behavior. The vehicle transaction characteristics mentioned above can include information on the car manufacturer, dealer, and the branch that issued the loan, which can be used to capture potential risk differences arising from different vehicle brands and transaction channels.

[0023] Step S220: Cluster the historical data to obtain multiple clustered datasets, each with different default risk characteristics; In this embodiment, the clustering described above can divide users into multiple groups with different risk characteristics, thus obtaining multiple clustered datasets with different default risk characteristics. This clustering can be achieved using existing clustering algorithms, such as the K-Means clustering algorithm, which will not be elaborated upon here. To improve the clustering effect and better reflect the differences in default risk characteristics among the clustered datasets, the clustering can also be achieved by first constructing composite features and then combining them with the original historical data for clustering.

[0024] In some embodiments, the clustering of the historical data to obtain multiple clustered datasets includes: First, based on the basic features in the loan data of each historical user, composite features for each historical user are constructed according to the preset composite feature construction rules. In this embodiment, the aforementioned preset composite feature construction rules can be predetermined according to actual needs, and can be used to combine one or more features from the basic features according to the composite feature construction rules to obtain composite features.

[0025] In some embodiments, specific composite features for auto loans can be constructed by combining the characteristics of auto loans to reflect factors such as debt burden, repayment behavior, credit activity, and application credibility. That is, the composite features for each historical user are constructed based on the basic features in the loan data of each historical user, according to preset composite feature construction rules, including: The first step is to construct a first composite feature and a second composite feature based on the loan characteristics and user characteristics in the loan data of each historical user. The first composite feature is used to assess the user's debt pressure and repayment ability, and the second composite feature is used to assess the user's historical credit. In this embodiment, the first composite feature can be a debt pressure and repayment ability feature. This type of feature can assess a user's overall debt level and repayment pressure, such as the overall debt-to-income ratio, loan-to-asset ratio, and outstanding debt coverage ratio, and can be used to quantify the applicant's repayment pressure. The first composite feature can be obtained by determining the overall debt-to-income ratio, loan-to-asset ratio, and outstanding debt coverage ratio based on loan characteristics and user characteristics respectively. The second composite feature can be a repayment behavior and historical performance feature. This type of feature can reflect a user's historical credit, such as overall delinquency intensity, recent default trends, account activity, and health, and can be used to assess the applicant's historical credit status. The second composite feature can be obtained by determining the overall delinquency intensity, recent default trends, account activity, and health based on loan characteristics and user characteristics respectively.

[0026] The second step is to construct a third composite feature based on the loan characteristics in the loan data of each historical user. The third composite feature is used to assess the frequency with which a user seeks a new loan. In this embodiment, the aforementioned third composite feature can be a credit activity and thirst feature. Such features can assess the frequency with which a user seeks new loans, such as recent credit inquiry density and credit utilization rate, and can identify applicants with tight funds. Alternatively, the third composite feature can be obtained by determining recent credit inquiry density, credit utilization rate, etc., based on loan characteristics.

[0027] The third step is to construct a fourth composite feature based on the user characteristics in each historical user loan data. The fourth composite feature is used to represent the completeness and stability of the user's application materials. In this embodiment, the fourth composite feature mentioned above can be an application credibility and stability feature. Such features can assess the completeness and stability of the user's application materials, such as identity verification strength and job stability.

[0028] The fourth step involves obtaining the composite features of each historical user based on the first composite feature, the second composite feature, the third composite feature, and the fourth composite feature.

[0029] In this embodiment, the first, second, third, and fourth composite features constitute a composite feature. Each historical user loan data point can be used to construct a composite feature through the above steps.

[0030] By constructing the first, second, third, and fourth composite features respectively, the characteristics of auto loans are incorporated, resulting in strong business interpretability. This allows the composite features of each historical user to better reflect the characteristics of auto loans, thereby helping to accurately predict loan defaults.

[0031] Then, based on the historical data and the composite features of each historical user, an initial training dataset is obtained; In this embodiment, each original historical user loan data set contains basic features and a corresponding loan default label. Combining these with composite features yields new historical user loan data sets, thus creating the initial training dataset. Each new historical user loan data set includes basic features, composite features, and a corresponding loan default label.

[0032] Finally, the initial training dataset is clustered to obtain multiple clustered datasets.

[0033] In this embodiment, the basic and composite features in the initial training dataset can be used as features strongly correlated with default risk. The clustering described above can be implemented using Gaussian mixture models, K-Means clustering algorithms, etc., which are existing technologies and will not be elaborated further here. Clustering can decompose the complex concept of default risk into multiple subgroups with distinct characteristics and interpretability.

[0034] In some embodiments, clustering the initial training dataset to obtain multiple clustered datasets includes: using the K-Means clustering algorithm to cluster the initial training dataset to obtain multiple clustered datasets.

[0035] In this embodiment, K-Means has relatively low computational complexity and can quickly process datasets containing tens of thousands or even hundreds of thousands of samples, thereby improving the clustering effect.

[0036] By constructing composite features for each historical user based on the basic features in their loan data according to pre-defined composite feature construction rules, an initial training dataset is obtained based on the historical data and the composite features of each historical user. Finally, the initial training dataset is clustered to achieve a clustering with clear business meaning. This can transform global oversampling into multiple local oversampling, and the generated minority class samples are more consistent with the data distribution of their respective groups, resulting in high business rationality and helping to improve the recall and AUC value of the default prediction model.

[0037] Step S230: Perform default risk assessment on each clustered dataset to obtain the risk level of each clustered dataset; In this embodiment, the default risk characteristics of different clustered datasets can be used to assess the risk level of each clustered dataset separately.

[0038] In some embodiments, the step of assessing the default risk of each clustered dataset to obtain the risk level of each clustered dataset includes: First, the historical default rate of each cluster dataset is calculated based on the loan default labels in each cluster dataset. In this embodiment, the loan default labels mentioned above can be pre-labeled based on actual circumstances. For example, `loan_default` can be used as the loan default label to indicate whether a user is overdue, where `loan_default=1` indicates that the user is overdue, and `loan_default=0` indicates that the user is not overdue. The historical default rate of the clustered dataset can be calculated by combining the number of loan default labels for each user in the clustered dataset.

[0039] In some embodiments, calculating the historical default rate of each cluster dataset based on the loan default labels in each cluster dataset includes: The first step is to calculate the number of defaulting users in each cluster dataset based on the loan default labels in each cluster dataset. The second step is to obtain the historical default rate of each cluster dataset based on the ratio of the number of defaulting users to the total number of users in each cluster dataset.

[0040] In this embodiment, taking loan_default as the loan default label as an example, the historical default rate of a certain cluster = (the number of all users in the cluster marked as loan_default=1) / (the total number of users in the cluster).

[0041] The historical default rate of each cluster dataset can be quickly and accurately represented by calculating the ratio of defaulting users to the total number of users in each cluster dataset.

[0042] Then, the historical default rate of each cluster dataset is compared with the preset default threshold to obtain the risk level of each cluster dataset.

[0043] In this embodiment, the historical default rate can be calculated based on the loan default labels in the clustered dataset. Then, the historical default rate is compared with a preset default threshold to obtain the risk level. The preset default threshold can be set in advance according to actual circumstances. By judging the magnitude of the historical default rate and the preset default threshold, the risk level can be quickly determined. For example, if the historical default rate is less than the preset default threshold, the risk level is low; if the historical default rate is not less than the preset default threshold, the risk level is high.

[0044] Step S240: Based on the risk level of each clustered dataset, determine the oversampling strategy for each clustered dataset; In this embodiment, different risk levels can correspond to different adoption strategies. For example, a conservative oversampling strategy is used for high-risk clustering, while an aggressive oversampling strategy is used for low-risk clustering.

[0045] Step S250: Based on the oversampling strategy of each cluster dataset, oversample each cluster dataset to obtain a balanced dataset; In this embodiment, oversampling is performed according to the oversampling strategy of each cluster dataset to obtain a balanced dataset.

[0046] In some embodiments, the oversampling strategy based on each clustered dataset, performing oversampling on each clustered dataset to obtain a balanced dataset, includes: First, based on the oversampling strategy of each cluster dataset, the oversampling factor of each cluster dataset is obtained; In this embodiment, different oversampling strategies can have different oversampling factors preset. For example, for a conservative oversampling strategy, a lower oversampling factor can be set to prevent the over-generation of high-risk samples and the introduction of noise; for an aggressive oversampling strategy, a higher oversampling factor can be set to fully learn the default patterns of its minority classes. It should be noted that in specific implementations, the oversampling factor can be set according to the actual situation, and this embodiment does not impose any limitations.

[0047] Then, based on the oversampling factor of each cluster dataset, the SMOTE algorithm is used to oversample each cluster dataset to obtain a balanced dataset.

[0048] In this embodiment, the oversampling factor represents how many new minority class samples need to be generated; different oversampling factors lead to drastically different data distributions. The Synthetic Minority Over-sampling Technique (SMOTE) creates new, reasonable minority class samples based on the oversampling factor, rather than simply copying them. It generates synthetic new samples by interpolating between existing minority class samples, thereby enriching and expanding the minority class data distribution, ultimately achieving a balanced dataset. The above-described oversampling of the various clustering datasets using the SMOTE algorithm can be implemented using existing techniques and will not be elaborated further here.

[0049] This refined, business-oriented oversampling strategy not only solves the data imbalance problem but also proactively shapes a training dataset with richer information and a clearer structure, thus laying the foundation for building a powerful and accurate default prediction model and helping to train a better default prediction model.

[0050] Step S260: Based on the balanced dataset, train the preset default prediction model to obtain a loan default prediction model for use in loan default prediction.

[0051] In this embodiment, the aforementioned pre-built default prediction model can be a Gradient Boosting Decision Tree (GBDT) model, such as extreme gradient boosting (XGBoost) or lightweight gradient boosting machine (LightGBM). After balancing the dataset through the above steps, a GBDT default prediction model is trained. Deploying the trained model online allows for automated default risk prediction of new auto loan applications.

[0052] In the above implementation process, historical data is acquired, including multiple historical user loan data sets. Each historical user loan data set includes basic features and a corresponding loan default label. The basic features include at least user features, loan features, and vehicle transaction features. The historical data is clustered to obtain multiple cluster datasets, each with different default risk characteristics. Default risk assessments are performed on each cluster dataset to obtain its risk level. Based on the risk levels of each cluster dataset, an oversampling strategy is determined. Based on the oversampling strategy, each cluster dataset is oversampled to obtain a balanced dataset. Based on the balanced dataset, a pre-set default prediction model is trained to obtain a loan default prediction model for loan default prediction. By introducing a risk adaptive mechanism, different oversampling intensities are used for clusters with different risks, avoiding the blind creation of users in high-risk groups, making the model more robust and conforming to the principle of prudent risk control. Within the framework of clustering and oversampling, domain feature engineering and risk adaptive strategies are used to address the issue of poor performance in specific business scenarios, thereby improving the effectiveness of loan default prediction.

[0053] Please refer to Figure 3 , Figure 3 This is a structural block diagram of a loan default prediction device provided in an embodiment of the present invention. This embodiment provides a loan default prediction device, including an acquisition module 410, a clustering module 420, an evaluation module 430, a determination module 440, an oversampling module 450, and a training module 460, wherein: The acquisition module 410 is used to acquire historical data, which includes multiple historical user loan data. Each historical user loan data includes basic characteristics and a corresponding loan default tag. The basic characteristics include at least: user characteristics, loan characteristics and vehicle transaction characteristics. Clustering module 420 is used to cluster the historical data to obtain multiple clustered datasets, each with different default risk characteristics; The assessment module 430 is used to assess the default risk of each clustered dataset separately and obtain the risk level of each clustered dataset. The determination module 440 is used to determine the oversampling strategy for each cluster dataset based on the risk level of each cluster dataset; The oversampling module 450 is used to oversample each cluster dataset according to the oversampling strategy of each cluster dataset to obtain a balanced dataset. Training module 460 is used to train a pre-set default prediction model based on the balanced dataset to obtain a loan default prediction model for use in loan default prediction.

[0054] Historical data is acquired through the acquisition module 410. This historical data includes multiple historical user loan data sets, each containing basic features and a corresponding loan default label. The basic features include at least user characteristics, loan characteristics, and vehicle transaction characteristics. The clustering module 420 clusters the historical data to obtain multiple cluster datasets, each with different default risk characteristics. The evaluation module 430 assesses the default risk of each cluster dataset to obtain its risk level. The determination module 440 determines the oversampling strategy for each cluster dataset based on its risk level. The oversampling module 450 oversamples each cluster dataset based on its oversampling strategy to obtain a balanced dataset. The training module 460 trains a pre-set default prediction model based on the balanced dataset to obtain a loan default prediction model for loan default prediction. By introducing a risk adaptive mechanism, different oversampling intensities are used for clusters with different risks, avoiding the blind creation of users in high-risk groups, making the model more robust and conforming to the principle of prudent risk control. Within the framework of clustering and oversampling, domain feature engineering and risk adaptive strategies are used to address the issue of poor performance in specific business scenarios, thereby improving the effectiveness of loan default prediction.

[0055] Please see Figure 4 , Figure 4 This is a schematic structural block diagram of an electronic device provided in an embodiment of this application. The electronic device includes a memory 101, a processor 102, and a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected to each other directly or indirectly to realize data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, such as the program instructions / modules corresponding to the loan default prediction device provided in this embodiment of the application. The processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used to communicate with other node devices for signaling or data.

[0056] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0057] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0058] Understandable. Figure 4 The structure shown is for illustrative purposes only; the electronic device may also include components that are more advanced than those shown. Figure 4 The more or fewer components shown, or having the same Figure 4 The different configurations shown. Figure 4 The components shown can be implemented using hardware, software, or a combination thereof.

[0059] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0060] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0061] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0063] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method of loan default prediction, characterized by, The method comprises the following steps: obtaining historical data, wherein the historical data comprises a plurality of historical user loan data, each historical user loan data comprises basic features and a corresponding loan default label, and the basic features at least include user features, loan features and vehicle transaction features; clustering the historical data to obtain a plurality of clustering data sets, wherein the default risk features of each clustering data set are different; respectively performing default risk assessment on each clustering data set to obtain a risk level of each clustering data set; determining a oversampling strategy of each clustering data set based on the risk level of each clustering data set; respectively performing oversampling on each clustering data set based on the oversampling strategy of each clustering data set to obtain a balanced data set; training a preset default prediction model based on the balanced data set to obtain a loan default prediction model for loan default prediction.

2. The loan default prediction method of claim 1, wherein, The clustering of the historical data to obtain a plurality of clustering data sets comprises: constructing a composite feature of each historical user based on the basic features in each historical user loan data according to a preset composite feature construction rule; obtaining an initial training data set based on the historical data and the composite feature of each historical user; clustering the initial training data set to obtain a plurality of clustering data sets.

3. The loan default prediction method of claim 2, wherein, The construction of the composite feature of each historical user based on the basic features in each historical user loan data according to the preset composite feature construction rule comprises: constructing a first composite feature and a second composite feature based on the loan features and the user features in each historical user loan data, wherein the first composite feature is used to evaluate the debt pressure and repayment ability of the user, and the second composite feature is used to evaluate the historical credit of the user; constructing a third composite feature based on the loan features in each historical user loan data, wherein the third composite feature is used to evaluate the frequency of the user seeking new loans; constructing a fourth composite feature based on the user features in each historical user loan data, wherein the fourth composite feature is used to represent the completeness and stability of the user's application materials; obtaining the composite feature of each historical user based on the first composite feature, the second composite feature, the third composite feature and the fourth composite feature.

4. The loan default prediction method of claim 2, wherein, The clustering of the initial training data set to obtain a plurality of clustering data sets comprises: clustering the initial training data set by using a K-Means clustering algorithm to obtain a plurality of clustering data sets.

5. The loan default prediction method of claim 1, wherein, The default risk assessment on each clustering data set to obtain a risk level of each clustering data set comprises: calculating a historical default rate of each clustering data set based on the loan default label in each clustering data set; comparing the historical default rate of each clustering data set with a preset default threshold to obtain a risk level of each clustering data set.

6. The loan default prediction method of claim 5, wherein, The historical default rates of the respective clustering data sets are calculated based on the loan default labels in the respective clustering data sets, respectively, including: The number of default users in the respective clustering data sets is calculated based on the loan default labels in the respective clustering data sets, respectively; The historical default rates of the respective clustering data sets are obtained based on the ratio of the number of default users to the total number of users in the respective clustering data sets, respectively.

7. The loan default prediction method of claim 1, wherein, The respective clustering data sets are oversampled based on the oversampling strategies of the respective clustering data sets to obtain balanced data sets, including: The oversampling multiples of the respective clustering data sets are obtained based on the oversampling strategies of the respective clustering data sets; The respective clustering data sets are oversampled based on the oversampling multiples of the respective clustering data sets using the SMOTE algorithm to obtain balanced data sets.

8. A loan default prediction apparatus characterized by comprising: including: The acquisition module is configured to acquire historical data, wherein the historical data includes a plurality of historical user loan data, each historical user loan data includes a basic feature and a corresponding loan default label, and the basic feature at least includes a user feature, a loan feature, and a vehicle transaction feature; The clustering module is configured to cluster the historical data to obtain a plurality of clustering data sets, wherein the default risk features of the respective clustering data sets are different; The evaluation module is configured to evaluate the default risk of each clustering data set to obtain the risk level of the respective clustering data sets; The determination module is configured to determine the oversampling strategy of each clustering data set based on the risk level of the respective clustering data sets; The oversampling module is configured to oversample the respective clustering data sets based on the oversampling strategies of the respective clustering data sets to obtain balanced data sets; The training module is configured to train a preset default prediction model based on the balanced data sets to obtain a loan default prediction model for loan default prediction.

Images