Training method of default prediction model, default prediction method, device and equipment

Abstract

The disclosure provides a default prediction model training method, a default prediction method, a device and equipment, which can be applied to big data analysis, financial technology and artificial intelligence field. The default prediction model training method comprises: performing feature extraction on the repayment information of each object in the resource repayment data set to generate a repayment vector corresponding to the object, the object including a default object and a non-default object; performing cluster analysis on the repayment vectors of the plurality of default objects to obtain a plurality of data clusters and a cluster centroid corresponding to each data cluster; sampling the plurality of cluster centroids to generate at least one centroid pair; generating a plurality of default vectors based on the centroid distance between the first centroid and the second centroid in each centroid pair; and training the default prediction model using the default vectors and the resource repayment data set when the proportion of the total number of the default vectors and the repayment vectors corresponding to the plurality of default objects to the number of the repayment vectors corresponding to the plurality of non-default objects is a preset proportion.

Description

Technical Field

[0001] This disclosure relates to the fields of big data analytics, fintech, and artificial intelligence, and in particular to a training method, a default prediction method, an apparatus, and a device for a default prediction model. Background Technology

[0002] In recent years, defaults have become increasingly common. Predicting defaults among customers with loan needs facilitates monitoring of potential defaulters and helps reduce the risk of bad debts. During the training of default prediction models, due to the significant disparity between the number of defaulting and non-defaulting samples, oversampling is employed to improve the imbalance ratio.

[0003] In realizing the present invention, the inventors discovered that the related technology has at least the following problems: because the oversampling process only randomly samples among the nearest neighbors of the sample, it is easy to generate the problem of sample marginalization, which in turn leads to low training accuracy of the default prediction model. Summary of the Invention

[0004] In view of the above problems, this disclosure provides training methods, default prediction methods, apparatus, devices, media and program products for default prediction models.

[0005] According to a first aspect of this disclosure, a method for training a default prediction model is provided, comprising: extracting features from the repayment information of each object in a resource repayment dataset to generate a repayment vector corresponding to the object, wherein the object includes defaulting objects and non-defaulting objects, and the repayment information includes attribute information, historical repayment information, and credit information of the object; performing cluster analysis on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and a cluster centroid corresponding to each data cluster; sampling the multiple cluster centroids to generate at least one centroid pair; generating multiple default vectors based on the centroid distance between the first centroid and the second centroid in each centroid pair; and training a default prediction model using the default vectors and the resource repayment dataset, provided that the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio.

[0006] According to embodiments of this disclosure, the method further includes: for each centroid pair, determining the number of group defaults for the centroid pair, wherein the number of group defaults is the sum of the number of repayment vectors in the data clusters corresponding to the first centroid and the second centroid in the centroid pair; determining the new data generation ratio corresponding to the centroid pair based on the number of group defaults and the total number of repayment vectors in the resource repayment data; and determining the number of new group data corresponding to the centroid pair based on a preset number of new data and the new data generation ratio.

[0007] According to embodiments of this disclosure, generating multiple default vectors based on the centroid distance between the first centroid and the second centroid in each centroid pair includes: determining the centroid distance between the first centroid and the second centroid for each centroid pair; generating a new centroid distance for the number of new data pairs based on the centroid distance using a random function; and generating a default vector for the number of new data pairs based on the new centroid distance and the first centroid or the second centroid.

[0008] According to embodiments of this disclosure, before performing cluster analysis on the repayment vectors of the multiple defaulting objects, the method further includes: preprocessing the resource repayment dataset to generate a new resource repayment dataset, and using the new resource repayment dataset to generate default vectors.

[0009] According to embodiments of this disclosure, the preprocessing of the resource repayment dataset includes: generating a resource table based on the attribute categories corresponding to the data in the resource repayment dataset; deleting the row data for each row in the resource table if the missing rate of the row data is greater than a first preset missing rate; deleting the column data for each column in the resource table if the missing rate of the column data is greater than a second preset missing rate; supplementing the column data based on the probability distribution of the column data if the missing rate of the column data is less than or equal to the second preset missing rate; and deleting the row data of each individual data point in the column data if the individual data point is not within the preset standard deviation range of the column data.

[0010] According to embodiments of this disclosure, the preprocessing of the resource repayment dataset further includes binning the data in the resource repayment dataset whose attribute category is a preset category to obtain binned data.

[0011] According to embodiments of this disclosure, the preprocessing of the resource repayment dataset further includes deleting the column data if the variance inflation coefficient of each column data is greater than a preset coefficient.

[0012] The second aspect of this disclosure provides a default prediction method, comprising: acquiring a dataset to be predicted, wherein the dataset includes attribute information, historical repayment information and credit information of an object; inputting the dataset to be predicted into a default prediction model trained by the training method of the default prediction model, and outputting the default probability of the object.

[0013] A third aspect of this disclosure provides a training apparatus for a default prediction model, comprising: a feature extraction module for extracting features from the repayment information of each object in a resource repayment dataset to generate a repayment vector corresponding to the object, wherein the object includes defaulting objects and non-defaulting objects, and the repayment information includes attribute information, historical repayment information, and credit information of the object; a clustering module for performing cluster analysis on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and a cluster centroid corresponding to each data cluster; a sampling module for sampling the multiple cluster centroids to generate at least one centroid pair; a data generation module for generating multiple default vectors based on the centroid distance between a first centroid and a second centroid in each centroid pair; and a training module for training the default prediction model using the default vectors and the resource repayment dataset, provided that the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio.

[0014] The fourth aspect of this disclosure provides a default prediction device, comprising: an acquisition module for acquiring a dataset to be predicted, the dataset including attribute information, historical repayment information and credit information of an object; and a prediction module for inputting the dataset to be predicted into a default prediction model trained by the training method of the default prediction model, and outputting the default probability of the object.

[0015] A fifth aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors perform the method described above.

[0016] A sixth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods described above.

[0017] The seventh aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0018] According to the training method, apparatus, device, medium, and program products of the default prediction model provided in this disclosure, multiple data clusters and their corresponding centroids are obtained by clustering the repayment vectors of multiple defaulting objects. The multiple centroids are sampled to generate at least one centroid pair. Based on the centroid distance between the first and second centroids in each centroid pair, multiple default vectors are generated. Since the default vectors are generated based on the centroid distance between centroid pairs, they fall within the range defined by the multiple cluster centroids, thus at least partially solving the problem of marginalization of newly generated data. This effectively changes the imbalanced state of the resource repayment dataset, resulting in higher accuracy for the default prediction model trained on the resource repayment dataset and default vectors. Attached Figure Description

[0019] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The diagram illustrates a training method for a default prediction model or an application scenario of a default prediction method according to embodiments of the present disclosure.

[0021] Figure 2 A flowchart illustrating a method for training a default prediction model according to an embodiment of the present disclosure is shown schematically.

[0022] Figure 3 A flowchart illustrating a default prediction method according to an embodiment of the present disclosure is shown schematically.

[0023] Figure 4 This schematic diagram illustrates a structural block diagram of a training apparatus for a default prediction model according to an embodiment of the present disclosure;

[0024] Figure 5 A schematic block diagram of a default prediction device according to an embodiment of the present disclosure is shown; and

[0025] Figure 6 A block diagram of an electronic device suitable for implementing a training method or default prediction method for a default prediction model according to embodiments of the present disclosure is shown schematically. Detailed Implementation

[0026] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0028] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0029] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0030] In the technical solutions disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of data (including but not limited to personal information of the subjects) comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0031] In recent years, defaults have become increasingly common. Predicting defaults among customers with loan needs allows for better monitoring of potential defaulters, helping to reduce the risk of bad debts. During the prediction process, because the number of defaulting samples in the dataset is far less than the number of non-defaulting samples, the default prediction model spends more effort learning from the non-defaulting samples. However, identifying the defaulting samples is actually more important. Therefore, there are two main approaches to handling imbalanced datasets: undersampling and oversampling.

[0032] The core of undersampling is to improve the imbalance ratio by reducing the number of majority class samples. For a default dataset, 100 samples are randomly selected from 1000 non-default samples and added to the training set along with the 100 default samples. It can be seen that undersampling discards a large portion of the non-default samples, thus losing a significant amount of information and easily leading to underfitting of the model.

[0033] The core of oversampling methods is to improve the imbalance ratio by increasing the number of minority class samples. In related techniques, oversampling can be broadly divided into two types: random oversampling and SMOTE (Synthetic Minority Over-sampling Technique).

[0034] Random oversampling involves repeatedly drawing from 100 default samples 900 times on a default dataset, combining them with the previous 100 samples to form 1000 default samples. This method results in duplicate samples, which can easily lead to model overfitting.

[0035] SMOTE oversampling calculates the distance from one defaulting sample to all other defaulting samples using Euclidean distance as the standard for each defaulting sample, resulting in multiple nearest neighbor samples. Then, a sampling scalar is set according to the imbalance ratio. For each defaulting sample, samples with the sampling scalar are randomly selected from the multiple nearest neighbor samples, and the defaulting sample is connected to the multiple randomly selected samples to generate new samples.

[0036] While the SMOTE algorithm effectively addresses the overfitting problem caused by duplicate samples, it cannot overcome the data distribution issues in imbalanced datasets, easily leading to data marginalization. When generating new defaulted samples, the SMOTE algorithm simply interpolates randomly among their nearest neighbors without considering the distribution of non-defaulted samples around the defaulted sample. This results in newly generated defaulted samples being classified as non-defaulted samples, thus exhibiting a degree of blind conformity.

[0037] In view of the above, embodiments of this disclosure provide a method for training a default prediction model, a default prediction method, a training device for a default prediction model, a default prediction device, an electronic device, a readable storage medium, and a computer program product. The method for training the default prediction model includes: extracting features from the repayment information of each object in a resource repayment dataset to generate a repayment vector corresponding to the object, wherein the object includes defaulting objects and non-defaulting objects, and the repayment information includes the object's attribute information, historical repayment information, and credit information; performing cluster analysis on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and a cluster centroid corresponding to each data cluster; sampling the multiple cluster centroids to generate at least one centroid pair; generating multiple default vectors based on the centroid distance between the first centroid and the second centroid in each centroid pair; and training the default prediction model using the default vectors and the resource repayment dataset, provided that the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio.

[0038] Figure 1 The diagram illustrates a training method for a default prediction model or an application scenario of a default prediction method according to an embodiment of the present disclosure.

[0039] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0040] An object can interact with the server 105 via the network 104 using at least one of the first terminal device 101, the second terminal device 102, and the third terminal device 103 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0041] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0042] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by objects using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received object requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to object requests) to the terminal devices.

[0043] It should be noted that the training method or default prediction method of the default prediction model provided in this disclosure embodiment can generally be executed by server 105. Correspondingly, the training device or default prediction device of the default prediction model provided in this disclosure embodiment can generally be located in server 105. The training method or default prediction method of the default prediction model provided in this disclosure embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the training device or default prediction device of the default prediction model provided in this disclosure embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0044] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0045] The following will be based on Figure 1 The described scene, through Figure 2 The training method of the default prediction model of the disclosed embodiments is described in detail.

[0046] Figure 2 A flowchart illustrating a training method for a default prediction model according to an embodiment of the present disclosure is shown.

[0047] like Figure 2 As shown, the training method for the default prediction model includes operations S210 to S250.

[0048] In operation S210, feature extraction is performed on the repayment information of each object in the resource repayment dataset to generate a repayment vector corresponding to the object. The objects include defaulting objects and non-defaulting objects, and the repayment information includes the object's attribute information, historical repayment information, and credit information.

[0049] In operation S220, cluster analysis is performed on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and the cluster centroids corresponding to each data cluster.

[0050] In operation S230, multiple cluster centroids are sampled to generate at least one centroid pair.

[0051] In operation S240, multiple default vectors are generated based on the centroid distance between the first and second centroids in each centroid pair.

[0052] In operation S250, with the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects being a preset ratio, the default prediction model is trained using default vectors and resource repayment datasets.

[0053] According to embodiments of this disclosure, the resource repayment dataset centrally includes a default dataset and a non-default dataset. The default dataset includes attribute information, historical repayment information, and credit information for multiple defaulting and non-defaulting entities. Attribute information includes names, ages, addresses, etc. Historical repayment information includes information such as the number of loan applications, the number of overdue payments, the outstanding loan balance, and the total approved loan amount for all loans. Credit information includes credit records and credit scores. It should be noted that the acquisition of attribute information, historical repayment information, and credit information complies with relevant laws and regulations, and necessary confidentiality measures have been taken. Furthermore, user authorization or consent was obtained before acquiring attribute information, historical repayment information, and credit information.

[0054] According to embodiments of this disclosure, feature extraction is performed on the attribute information, historical repayment information, and credit information of each object in the resource repayment dataset to generate a repayment vector corresponding to each object. Cluster analysis is then performed on the repayment vectors of the defaulting objects. During the cluster analysis, k-means clustering algorithm can be used to obtain multiple data clusters and the centroid of each cluster.

[0055] According to embodiments of this disclosure, in the process of cluster analysis using k-means, K points are first randomly selected as cluster centroids. The distances from other vectors in the repayment vector to the cluster centroids are calculated, and these vectors are assigned as the categories of the nearest data clusters. The average value of each category is determined for the classified repayment vectors, and new cluster centroids are calculated. The new cluster centroids are compared with the previous cluster centroids. If the cluster centroids change, the distances from other vectors in the repayment vector to the cluster centroids are calculated iteratively again, and new cluster centroids are calculated, until the cluster centroids no longer change, thus completing the calculation of the cluster centroids.

[0056] According to embodiments of this disclosure, multiple data clusters can be represented as F = {F1, F2, F3, ..., F...}n Each repayment vector within a data cluster is represented as F. i ={x i1 ,x i2 ,x i3 ,...,x in Each data cluster consists of n compensation vectors. The average of these n compensation vectors is used to obtain the cluster centroid FC of the data cluster. i The final set of cluster centroids is FC = {FC1, FC2, FC3, ..., FC...} n Cluster centroid FC of data clusters i The calculation formula (1) is as follows:

[0057]

[0058] According to embodiments of this disclosure, the set of cluster centroids is FC = {FC1, FC2, FC3, ..., FC...} n Randomly sample multiple cluster centroids and pair them up to generate at least one centroid pair. Connect the first and second centroids in each centroid pair to determine the centroid distance between the pairs. For example, if there are a total of 4 data clusters, the pairing result would be (FC1, FC2) and (FC3, FC4).

[0059] According to embodiments of this disclosure, multiple default vectors are generated on the line connecting the first and second centroids based on the centroid distance. When the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio, the default prediction model is trained using default vectors and a resource repayment dataset. The preset ratio is set based on the number of repayment vectors for defaulting objects and the number of repayment vectors for non-defaulting objects, and corresponding ratios can be set for different types of resource repayment datasets. Specifically, the preset ratio does not need to reach 1, because the number of defaulting objects is inherently smaller than the number of non-defaulting objects, and while changing the degree of imbalance, it is necessary to maintain the bias of the data. For example, the preset ratio can be set to 4 / 6. The calculation formula (2) for the preset ratio is as follows:

[0060]

[0061] Where IMR represents the preset ratio, N min N represents the number of repayment vectors for defaulting entities. max This represents the number of repayment vectors for non-defaulting objects.

[0062] According to embodiments of this disclosure, default vectors and repayment vectors corresponding to defaulting and non-defaulting objects in the resource dataset are input into a random forest model for training, resulting in a trained default prediction model. Random forest is a combined classifier using decision trees as base classifiers. It employs the bagging (bootstrap aggregating) technique from ensemble learning. The bagging technique allows each tree to be processed in parallel, resulting in faster training speed and better accuracy. During the construction of the random forest, samples and features are randomly selected each time, improving the default prediction model's robustness and generalization ability.

[0063] According to embodiments of this disclosure, multiple data clusters and their corresponding centroids are obtained by clustering the repayment vectors of multiple defaulting objects. The centroids are sampled to generate at least one centroid pair. Multiple default vectors are generated based on the centroid distance between the first and second centroids in each centroid pair. Since the default vectors are generated based on the centroid distance between centroid pairs, they fall within the bounded area of ​​the multiple cluster centroids, thus at least partially solving the problem of marginalization of newly generated data. This effectively changes the imbalanced state of the resource repayment dataset, resulting in higher accuracy for the default prediction model trained on the resource repayment dataset and default vectors.

[0064] According to embodiments of this disclosure, the training method for the default prediction model may further include the following operations:

[0065] For each centroid pair, determine the number of group defaults for the centroid pair. The number of group defaults is the sum of the number of repayment vectors in the data clusters corresponding to the first and second centroids of the centroid pair, respectively. Determine the new data generation ratio corresponding to the centroid pair based on the number of group defaults and the total number of repayment vectors in the resource repayment data. Determine the number of new data pairs corresponding to the centroid pair based on the preset number of new data and the new data generation ratio.

[0066] According to the embodiments of this disclosure, assuming the centroid pair is (FC1, FC2), the first centroid is FC1, and the second centroid is FC2. The number of repayment vectors in the data cluster corresponding to the first centroid and the number of repayment vectors in the data cluster corresponding to the second centroid are determined, and the two are added together to determine the number of group defaults. The new data generation ratio corresponding to the centroid pair is determined based on the number of group defaults and the total number of repayment vectors in the resource repayment data. The preset number of new data is multiplied by the new data generation ratio to obtain the number of new data for the group corresponding to the centroid pair, ensuring that the number of subsequently generated default vectors remains within its own data cluster. For example, if the preset number of new data is set to 50 and the new data generation ratio is 1 / 5, then the number of new data for the group is determined to be 10. The specific calculation formula (3) for the new data generation ratio is as follows:

[0067]

[0068] Where C represents the proportion of new data generated for the centroid pair (FC1, FC2), FC all This represents the total amount of repayment vectors in the resource repayment dataset.

[0069] According to embodiments of this disclosure, generating multiple default vectors based on the centroid distance between the first and second centroids in each centroid pair may include the following operations:

[0070] For each centroid pair, determine the centroid distance between the first and second centroids; use a random function to generate a new centroid distance based on the centroid distance to form new data pairs; generate a default vector for the new data pairs based on the new centroid distance and the first or second centroid.

[0071] According to embodiments of this disclosure, the first and second centroids in a centroid pair are connected to determine the centroid distance between the centroid pairs. A default vector is generated by connecting the centroids, ensuring that the default vector remains within the cluster centroids and reducing the likelihood of the default vector being divided into repayment vectors corresponding to non-defaulting objects. The formula (4) for calculating the default vector is as follows:

[0072] x new =FC y +rand(0,1)|FC y -FC z | (4)

[0073] Where, x new Denotes the default vector, FC y and FC z These represent one of the first and second centroids, respectively, and rand represents a random function.

[0074] According to embodiments of this disclosure, before performing cluster analysis on the repayment vectors of multiple defaulting entities, the following operations may also be included:

[0075] The resource repayment dataset is preprocessed to generate a new resource repayment dataset, which is then used to generate default vectors.

[0076] According to embodiments of this disclosure, the resource repayment dataset is preprocessed to make the data more accurate. Specifically, outliers are removed and missing values ​​are added, making the data more complete and concise, reducing the data volume and accelerating training. The processed data is then used for feature extraction and vectorization, allowing it to be input into the default prediction model for feature learning.

[0077] According to embodiments of this disclosure, preprocessing the resource repayment dataset may include the following operations:

[0078] A resource table is generated based on the attribute categories corresponding to the centralized data in the resource repayment dataset. For each row of data in the resource table, if the missing rate of the row data is greater than a first preset missing rate, the row data is deleted. For each column of data in the resource table, if the missing rate of the column data is greater than a second preset missing rate, the column data is deleted. If the missing rate of the column data is less than or equal to the second preset missing rate, the column data is supplemented based on the probability distribution of the column data. For each individual data in the column data, if the individual data is not within the preset standard deviation range of the column data, the row data of the row containing the individual data is deleted.

[0079] According to embodiments of this disclosure, a resource table is generated based on the attribute categories corresponding to the centralized data in the resource repayment dataset. Attribute categories may include name, outstanding loan balance, and credit score, etc. For missing values, the missing rate of the resource table needs to be statistically analyzed in both row and column directions. From a row perspective, if the missing rate of a row is greater than 60%, that row is deleted; from a column perspective, if the missing rate of a column is greater than 40%, that column is deleted.

[0080] According to embodiments of this disclosure, the retained data is supplemented by using the average value as filler for non-time series data, such as average loan term. For non-numerical series data, the mode and median are used as filler values, such as gender. For time series data, interpolation is used, and after sorting by time, the median value between the two values ​​is taken as supplementary data.

[0081] According to embodiments of this disclosure, outliers are tested using the mean and variance. In statistics, if a data distribution is approximately normal, then approximately 68% of the data will be within one standard deviation of the mean, approximately 95% will be within two standard deviations, and approximately 99.7% will be within three standard deviations. Since it is necessary to retain as much data as possible, a three-standard-deviation range is used. For each individual data point in a column, if the individual data point is not within the three-standard-deviation range of the column data, the row containing that individual data point is deleted.

[0082] According to embodiments of this disclosure, preprocessing the resource repayment dataset further includes the following operations:

[0083] Data in the resource repayment dataset that belongs to a preset attribute category is binned to obtain binned data.

[0084] According to embodiments of this disclosure, in order to reduce the complexity of the default prediction model and minimize the impact of variable noise on the model, the correlation between independent and dependent variables is increased to make the model more stable. Binning is employed for feature transformation, primarily binning monetary data, such as the total loan amount disbursed and the total loan amount approved for all loans in the account.

[0085] According to embodiments of this disclosure, preprocessing the resource repayment dataset further includes the following operations:

[0086] If the variance inflation coefficient of each column of data exceeds a preset coefficient, delete the column data.

[0087] According to embodiments of this disclosure, the variance inflation factor (VIF) is used to analyze the relationship between features of multiple columns of data. A higher VIF value indicates a better fit. If the VIF value is greater than a preset coefficient, it indicates that the features of that column of data are more collinear with other features, and therefore can be removed.

[0088] Figure 3 A flowchart illustrating a default prediction method according to an embodiment of the present disclosure is shown schematically.

[0089] like Figure 3 As shown, the default prediction method includes operation S310 and operation S320.

[0090] In operation S310, the dataset to be predicted is obtained. The dataset to be predicted includes the object's attribute information, historical repayment information, and credit information.

[0091] In operation S320, the default prediction model is trained by inputting the dataset to be predicted into the default prediction model and outputting the default probability of the object.

[0092] According to embodiments of this disclosure, attribute information, historical repayment information, and credit information of an object are obtained, and feature transformation is performed on the dataset to be predicted to generate a repayment vector. It should be noted that the acquisition of attribute information, historical repayment information, and credit information complies with relevant laws and regulations, and necessary confidentiality measures are taken. Furthermore, user authorization or consent is obtained before acquiring attribute information, historical repayment information, and credit information.

[0093] According to embodiments of this disclosure, a predicted repayment vector is input into a default prediction model. Since the default prediction model is built based on a random forest algorithm, where each decision tree is a classifier, for the input predicted repayment vector, N trees will produce N classification results. The random forest integrates the voting results of all classifiers, designating the category with the most votes as the final output. The categories include default and non-default classes, and the category to which the predicted dataset belongs is determined probabilistically. By performing default prediction, attention can be strengthened to potential defaulters, helping to reduce the risk of bad debts.

[0094] Based on the training method of the default prediction model described above, this disclosure also provides a training device for the default prediction model. The following will combine... Figure 4 The device is described in detail.

[0095] Figure 4 A schematic block diagram of a training apparatus for a default prediction model according to an embodiment of the present disclosure is shown.

[0096] like Figure 4 As shown, the training device 400 for the default prediction model in this embodiment includes a feature extraction module 410, a clustering module 420, a sampling module 430, a data generation module 440, and a training module 450.

[0097] The feature extraction module 410 is used to extract features from the repayment information of each object in the resource repayment dataset and generate a repayment vector corresponding to the object. The objects include defaulting objects and non-defaulting objects, and the repayment information includes the object's attribute information, historical repayment information, and credit information. In one embodiment, the feature extraction module 410 can be used to perform the operation S210 described above, which will not be repeated here.

[0098] Clustering module 420 is used to perform cluster analysis on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and the cluster centroid corresponding to each data cluster. In one embodiment, clustering module 420 can be used to perform the operation S220 described above, which will not be repeated here.

[0099] The sampling module 430 is used to sample multiple cluster centroids to generate at least one centroid pair. In one embodiment, the sampling module 430 can be used to perform the operation S230 described above, which will not be repeated here.

[0100] The data generation module 440 is used to generate multiple default vectors based on the centroid distance between the first centroid and the second centroid in each centroid pair. In one embodiment, the data generation module 440 can be used to perform the operation S240 described above, which will not be repeated here.

[0101] The training module 450 is used to train a default prediction model using default vectors and a resource repayment dataset, provided that the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio. In one embodiment, the training module 450 can be used to perform the operation S250 described above, which will not be repeated here.

[0102] According to embodiments of this disclosure, multiple data clusters and their corresponding centroids are obtained by clustering the repayment vectors of multiple defaulting objects. The centroids are sampled to generate at least one centroid pair. Multiple default vectors are generated based on the centroid distance between the first and second centroids in each centroid pair. Since the default vectors are generated based on the centroid distance between centroid pairs, they fall within the bounded area of ​​the multiple cluster centroids, thus at least partially solving the problem of marginalization of newly generated data. This effectively changes the imbalanced state of the resource repayment dataset, resulting in higher accuracy for the default prediction model trained on the resource repayment dataset and default vectors.

[0103] According to embodiments of this disclosure, the training apparatus 400 for the default prediction model further includes a default quantity determination module, a proportion determination module, and a new data determination module.

[0104] The default quantity determination module is used to determine the group default quantity for each centroid pair. The group default quantity is the sum of the number of repayment vectors in the data clusters corresponding to the first and second centroids of the centroid pair, respectively.

[0105] The proportion determination module is used to determine the proportion of new data generation corresponding to the centroid pair based on the number of team defaults and the total amount of repayment vectors in the resource repayment dataset.

[0106] The new data determination module is used to determine the number of new data pairs corresponding to the centroid pairs based on the preset number of new data pairs and the new data generation ratio.

[0107] According to embodiments of this disclosure, the data generation module 440 includes a first distance determination submodule, a second distance determination submodule, and a vector generation submodule.

[0108] The first distance determination submodule is used to determine the centroid distance between the first centroid and the second centroid for each centroid pair.

[0109] The second distance determination submodule is used to generate a new centroid distance based on the centroid distance using a random function to determine the number of new data in the team.

[0110] The vector generation submodule is used to generate a default vector for the number of new data points based on the new centroid distance and the first or second centroid.

[0111] According to embodiments of this disclosure, the training apparatus 400 for the default prediction model further includes a preprocessing module.

[0112] The preprocessing module is used to preprocess the resource repayment dataset to generate a new resource repayment dataset, which is then used to generate a default vector.

[0113] According to embodiments of this disclosure, the preprocessing module includes a resource table generation submodule, a first row deletion submodule, a first column deletion submodule, a supplementation module, and a second row deletion submodule.

[0114] The resource table generation submodule is used to generate resource tables based on the attribute categories corresponding to the central data in the resource repayment dataset.

[0115] The first row deletion submodule is used to delete row data for each row in the resource table if the missing data rate of the row data is greater than a first preset missing rate.

[0116] The first column deletion submodule is used to delete column data for each column of the resource table if the missing data rate of the column data is greater than the second preset missing rate.

[0117] The supplementation module is used to supplement column data based on the probability distribution of the column data when the column data is less than or equal to the second preset missing rate.

[0118] The second row deletion submodule is used to delete the row data of each individual data point in the column data if the individual data point is not within the preset standard deviation range of the column data.

[0119] According to embodiments of this disclosure, the preprocessing module further includes a binning module.

[0120] The binning module is used to bin data whose attribute category in the resource repayment dataset is a preset category, and obtain binned data.

[0121] According to embodiments of this disclosure, the preprocessing module further includes a second column deletion module.

[0122] The second column deletion module is used to delete column data when the variance inflation coefficient of each column data is greater than a preset coefficient.

[0123] Based on the above-described default prediction method, this disclosure also provides a default prediction device. The following will be combined with... Figure 5 The device is described in detail.

[0124] like Figure 5 As shown, the default prediction device 500 of this embodiment includes an acquisition module 510 and a prediction module 520.

[0125] The acquisition module 510 is used to acquire the dataset to be predicted, which includes the object's attribute information, historical repayment information, and credit information. In one embodiment, the feature acquisition module 510 can be used to perform the operation S310 described above, which will not be repeated here.

[0126] The prediction module 520 is used to input the dataset to be predicted into the default prediction model trained by the training method of the default prediction model, and output the default probability of the object. In one embodiment, the prediction module 520 can be used to perform the operation S320 described above, which will not be repeated here.

[0127] According to embodiments of this disclosure, any multiple modules among the feature extraction module 410, clustering module 420, sampling module 430, data generation module 440, and training module 450, or acquisition module 510 and prediction module 520, can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the feature extraction module 410, clustering module 420, sampling module 430, data generation module 440, and training module 450, or acquisition module 510 and prediction module 520, can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), programmable logic array (PLA), system-on-a-chip, system-on-a-substrate, system-on-package, application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the feature extraction module 410, clustering module 420, sampling module 430, data generation module 440, and training module 450, or acquisition module 510 and prediction module 520, can be at least partially implemented as a computer program module that can perform corresponding functions when the computer program module is run.

[0128] Figure 6 A block diagram of an electronic device suitable for implementing a training method or default prediction method for a default prediction model according to embodiments of the present disclosure is shown schematically.

[0129] like Figure 6As shown, an electronic device 600 according to an embodiment of this disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this disclosure.

[0130] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0131] According to embodiments of this disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0132] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0133] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0134] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the training method or default prediction method of the default prediction model provided in the embodiments of this disclosure.

[0135] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0136] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0137] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0138] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the object computing device, partially on the object device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the object computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0139] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. 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 indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated 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 or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0140] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0141] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A training method for a default prediction model, comprising: Feature extraction is performed on the repayment information of each object in the resource repayment dataset to generate a repayment vector corresponding to the object. The object includes defaulting objects and non-defaulting objects. The repayment information includes the object's attribute information, historical repayment information, and credit information. Cluster analysis is performed on the repayment vectors of multiple defaulting entities to obtain multiple data clusters and the cluster centroids corresponding to each data cluster. Sample multiple cluster centroids and pair them up to generate at least one centroid pair; For each centroid pair, determine the number of group defaults for the centroid pair. The number of group defaults is the sum of the number of repayment vectors in the data clusters corresponding to the first centroid and the second centroid in the centroid pair, respectively. The proportion of new data generation corresponding to the centroid pair is determined based on the number of team defaults and the total amount of repayment vectors in the resource repayment dataset. The number of new data points for the centroid pair is determined based on the preset number of new data points and the new data generation ratio. Based on the centroid distance between the first centroid and the second centroid in each centroid pair, a default vector for the number of new data in the group is generated on the line connecting the first centroid and the second centroid. When the ratio of the total number of default vectors and repayment vectors corresponding to multiple defaulting objects to the number of repayment vectors corresponding to multiple non-defaulting objects is a preset ratio, the default prediction model is trained using the default vectors and the resource repayment dataset.

2. The method according to claim 1, wherein, The method generates multiple default vectors based on the centroid distance between the first and second centroids in each centroid pair, including: For each of the said centroid pairs, determine the centroid distance between the first centroid and the second centroid; A new centroid distance is generated based on the centroid distance using a random function to generate the new number of data in the group; Based on the new centroid distance and the first centroid or the second centroid, generate the default vector of the number of new data in the team.

3. The method according to claim 1, wherein, Before performing cluster analysis on the repayment vectors of the multiple defaulting entities, the method further includes: The resource repayment dataset is preprocessed to generate a new resource repayment dataset, which is then used to generate a default vector.

4. The method according to claim 3, wherein, The preprocessing of the resource repayment dataset includes: A resource table is generated based on the attribute categories corresponding to the data in the resource repayment dataset. For each row of data in the resource table, if the missing rate of the row data is greater than a first preset missing rate, the row data is deleted. For each column of data in the resource table column direction, if the missing rate of the column data is greater than a second preset missing rate, the column data is deleted; If the column data is less than or equal to the second preset missing rate, the column data is supplemented based on the probability distribution of the column data; For each individual data point in the column data, if the individual data point is not within the preset standard deviation range of the column data, the row data of the row containing the individual data point is deleted.

5. The method according to claim 3, wherein, The preprocessing of the resource repayment dataset further includes: The data in the resource repayment dataset that belong to a preset category are binned to obtain binned data.

6. The method according to claim 4, wherein, The preprocessing of the resource repayment dataset further includes: If the variance inflation coefficient of each column of data is greater than a preset coefficient, delete the column of data.

7. A method for predicting default, comprising: Obtain the dataset to be predicted, which includes the object's attribute information, historical repayment information, and credit information; The default prediction model is trained by the default prediction model training method of any one of claims 1 to 6, and the default probability of the object is output.

8. A training device for a default prediction model, comprising: The feature extraction module is used to extract features from the repayment information of each object in the resource repayment dataset and generate a repayment vector corresponding to the object. The object includes defaulting objects and non-defaulting objects, and the repayment information includes the object's attribute information, historical repayment information, and credit information. The clustering module is used to perform clustering analysis on the repayment vectors of multiple defaulting objects to obtain multiple data clusters and the cluster centroids corresponding to each data cluster. The sampling module is used to sample multiple cluster centroids and pair them up to generate at least one centroid pair. The default quantity determination module is used to determine the group default quantity of each centroid pair, wherein the group default quantity is the sum of the number of repayment vectors in the data clusters corresponding to the first centroid and the second centroid in the centroid pair, respectively. The proportion determination module is used to determine the new data generation proportion corresponding to the centroid pair based on the number of team defaults and the total amount of repayment vectors in the resource repayment dataset; The new data determination module is used to determine the number of new data pairs corresponding to the centroid pairs based on a preset number of new data pairs and the new data generation ratio; the data generation module is used to generate a default vector of the number of new data pairs based on the centroid distance between the first centroid and the second centroid in each centroid pair, on the line connecting the first centroid and the second centroid. The training module is used to train the default prediction model using the default vector and the resource repayment dataset, provided that the ratio of the total number of default vectors and repayment vectors corresponding to the multiple defaulting objects to the number of repayment vectors corresponding to the multiple non-defaulting objects is a preset ratio.

9. A default prediction device, comprising: The acquisition module is used to acquire the dataset to be predicted, which includes the object's attribute information, historical repayment information, and credit information. The prediction module is used to input the dataset to be predicted into the default prediction model trained by the training method of the default prediction model according to any one of claims 1 to 6, and output the default probability of the object.

10. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 6, or claim 7.

11. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 6, or claim 7.

12. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 6, or claim 7.

Images