Individual repayment-free lending reviewing method and device based on decision tree algorithm

Abstract

The invention discloses a personal repayment-free lending auditing method and device based on a decision-making tree algorithm, and the method comprises the steps: obtaining and preprocessing the basic information of a customer, the credit record, and the comprehensive information of a data customer in the past, and carrying out the dimension reduction and model training through employing the decision-making tree algorithm; therefore, a decision model with good universality is constructed. The credit continuing qualification of the customer can be accurately evaluated, the working efficiency of the auditing process can be effectively improved, and the credit continuing process is accelerated.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method and device for reviewing personal interest-only loan renewals based on a decision tree algorithm. Background Art

[0002] The personal non-repayment loan renewal business can effectively solve the problem of capital turnover for individual private owners. However, in actual operation, due to changes in the market environment, relevant policies or the borrower's own conditions, the loan business will also experience twists and turns, risks need to be re-evaluated, loan amounts may need to be re-approved, and loan interest rates may also change. Currently, this process mainly relies on manual intervention, investigation, and changes to loan elements after manual approval. However, as the business volume increases, manual review intervention is time-consuming and laborious. Summary of the invention

[0003] This application provides a method and device for personal non-repayment loan renewal review based on a decision tree algorithm, which realizes automatic loan renewal business review by means of a decision tree algorithm, which helps to speed up the approval process and improve the accuracy. The technical solution is as follows.

[0004] In a first aspect, a method for reviewing personal principal-free loan renewal based on a decision tree algorithm is provided, the method comprising:

[0005] Obtain a customer information data set, wherein the customer information data set includes n-dimensional features of each customer among m customers, and the n-dimensional features of each customer include basic information of the customer, the credit record of the customer, and the past transaction record of the customer;

[0006] Preprocessing the customer information data set to obtain a preprocessed customer information data set;

[0007] Using the PCA algorithm to reduce the dimension of the preprocessed customer information data set to obtain a customer information data set after dimension reduction;

[0008] Based on the CART algorithm, a decision tree is trained using a target data set, wherein the target data set includes the customer information data set after dimensionality reduction and a loan renewal evaluation result for each of the m customers;

[0009] In response to receiving a loan renewal application request from a target customer, a loan renewal evaluation result of the target customer is determined based on information of the target customer using a trained decision tree, wherein the information of the target customer includes features of the n dimensions of the target customer.

[0010] In some implementations, preprocessing the customer information dataset to obtain a preprocessed customer information dataset includes:

[0011] For the features of each dimension in the n dimensions, determine the missing value ratio of the features of the dimension, where the missing value ratio represents the ratio between the number of customers who lack the features of the dimension among the m customers in the customer information dataset and m;

[0012] If the missing value ratio of the feature of the dimension exceeds the missing value ratio threshold, delete the feature of the dimension from the customer information data set; if the missing value ratio of the feature of the dimension does not exceed the missing value ratio threshold, use the average value of the feature of the dimension of m customers to fill the missing values ​​of the feature of the dimension of m customers;

[0013] The features of each dimension in the customer information dataset are normalized after deleting features and filling missing values.

[0014] In some implementations, the using of the PCA algorithm to reduce the dimension of the preprocessed customer information dataset to obtain a dimension-reduced customer information dataset includes:

[0015] For the feature of each dimension in the preprocessed customer information data set, based on the average value of the feature of the dimension and the standard deviation of the feature of the dimension, the feature of the dimension is standardized to obtain a standardized feature;

[0016] Determine the covariance matrix between the features of each dimension in the features of the n dimensions;

[0017] Decomposing the covariance matrix to obtain eigenvalues ​​and eigenvectors;

[0018] Select eigenvectors corresponding to the first p eigenvalues ​​from the eigenvalues, the sum of the p eigenvalues ​​being greater than a contribution rate threshold, and use the selected p eigenvectors as a customer information data set after dimensionality reduction.

[0019] In some embodiments, the method of training a decision tree using a target data set based on a CART algorithm comprises:

[0020] Based on the target data set, determine a target segmentation point of a feature of each dimension of the features of the n dimensions;

[0021] Recursively constructing a subtree based on the target segmentation point to obtain an initial decision tree;

[0022] Pruning the initial decision tree to obtain a pruned decision tree;

[0023] The hyperparameters in the pruned decision tree are adjusted based on the test set to obtain the trained decision tree.

[0024] In some implementations, determining a target segmentation point of a feature of each dimension of the features in the n dimensions based on the target data set includes:

[0025] For each dimension of the n-dimensional features and the loan renewal assessment result, multiple candidate split points are determined;

[0026] Based on the candidate segmentation points, the target data set is divided into two subsets, and the impurity or variance after segmentation is determined;

[0027] Select the split point that reduces the most impurity or variance to get the target split point.

[0028] In some implementations, pruning the initial decision tree to obtain a pruned decision tree includes:

[0029] For each node in the initial decision tree, determining the impurity of the segmentation after pruning the subtree of the node and the impurity of the segmentation of the subtree without pruning the node;

[0030] If the impurity of the segmentation is reduced after pruning the subtree of the node, the subtree of the node is pruned from the initial decision tree.

[0031] In a second aspect, a device for reviewing personal principal-free loan renewal based on a decision tree algorithm is provided, the device comprising:

[0032] An acquisition unit is used to acquire a customer information data set, wherein the customer information data set includes n-dimensional features of each customer among m customers, and the n-dimensional features of each customer include basic information of the customer, the credit record of the customer, and the past transaction record of the customer;

[0033] A preprocessing unit, used to preprocess the customer information data set to obtain a preprocessed customer information data set;

[0034] A dimension reduction unit, used for reducing the dimension of the preprocessed customer information data set by using a PCA algorithm to obtain a dimension-reduced customer information data set;

[0035] A training unit, configured to train a decision tree using a target data set based on a CART algorithm, wherein the target data set includes the customer information data set after dimensionality reduction and a loan renewal evaluation result for each of the m customers;

[0036] A determination unit is used to, in response to receiving a loan renewal application request from a target customer, determine a loan renewal evaluation result for the target customer based on information of the target customer using a trained decision tree, wherein the information of the target customer includes features of the target customer in the n dimensions.

[0037] In some embodiments, the preprocessing unit is used to determine the missing value ratio of the features of each dimension in n dimensions, the missing value ratio representing the ratio between the number of customers who are missing the features of the dimension among the m customers in the customer information dataset and m; if the missing value ratio of the features of the dimension exceeds a missing value ratio threshold, delete the features of the dimension from the customer information dataset; if the missing value ratio of the features of the dimension does not exceed the missing value ratio threshold, fill the missing values ​​of the features of the m customers in the dimension with the average value of the features of the m customers in the dimension; and normalize the features of each dimension in the customer information dataset after deleting the features and filling the missing values.

[0038] In some embodiments, the dimension reduction unit is used to standardize the features of each dimension in the preprocessed customer information data set based on the average value of the features of the dimension and the standard deviation of the features of the dimension to obtain standardized features; determine the covariance matrix between the features of each dimension among the n dimensional features; decompose the covariance matrix to obtain eigenvalues ​​and eigenvectors; select eigenvectors corresponding to the first p eigenvalues ​​from the eigenvalues, the sum of the p eigenvalues ​​being greater than a contribution rate threshold, and use the selected p eigenvectors as the customer information data set after dimensionality reduction.

[0039] In some embodiments, the training unit is used to determine the target segmentation point of the features of each dimension of the features of the n dimensions based on the target data set; recursively construct a subtree based on the target segmentation point to obtain an initial decision tree; prune the initial decision tree to obtain a pruned decision tree; and adjust the hyperparameters in the pruned decision tree based on a test set to obtain a trained decision tree.

[0040] In some embodiments, the training unit is used to determine multiple candidate segmentation points for the features of each dimension of the n-dimensional features and the renewal loan evaluation results; based on the candidate segmentation points, the target data set is divided into two subsets to determine the impurity or variance after segmentation; and the segmentation point that reduces the most impurity or variance is selected to obtain the target segmentation point.

[0041] In some embodiments, the training unit is used to determine, for each node in the initial decision tree, the impurity of the segmentation after pruning the subtree of the node and the impurity of the segmentation without pruning the subtree of the node; if the impurity of the segmentation is reduced after pruning the subtree of the node, the subtree of the node is pruned from the initial decision tree.

[0042] In a third aspect, a computer device is provided, which has the function of implementing the first aspect or any optional method of the first aspect. The computer device includes at least one unit, and the at least one unit is used to implement the method provided by the first aspect or any optional method of the first aspect.

[0043] In some embodiments, the unit in the computer device is implemented by software, and the unit in the computer device is a program module. In other embodiments, the unit in the computer device is implemented by hardware or firmware. The specific details of the computer device provided in the third aspect can be found in the first aspect or any optional manner of the first aspect, which will not be repeated here.

[0044] In a fourth aspect, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium. When the instruction is executed on a computer, the computer executes the method provided by the first aspect or any optional manner of the first aspect.

[0045] In a fifth aspect, a computer program product is provided, which includes one or more computer program instructions. When the computer program instructions are loaded and executed by a computer, the computer executes the method provided by the first aspect or any optional manner of the first aspect.

[0046] The embodiments of the present application have the following beneficial effects:

[0047] The embodiment of the present application proposes a method for reviewing personal principal-free loan renewals based on a decision tree algorithm. The method obtains and pre-processes the comprehensive information of the customer, such as the customer's basic information, credit record, past bank flow, etc., and uses a decision tree algorithm for dimensionality reduction and model training to construct a decision model with good versatility. This can not only accurately assess the customer's loan renewal qualifications, but also effectively improve the efficiency of the review process and speed up the loan renewal process, thereby partially or completely replacing the traditional manual review process. Through this method, financial institutions can manage loan renewal business more accurately and efficiently, while reducing potential credit risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the input parameters and output results of a decision tree algorithm provided in an embodiment of the present application;

[0049] Figure 2 It is a flow chart of a method for reviewing a personal non-principal repayment renewal loan based on a decision tree algorithm provided in an embodiment of the present application;

[0050] Figure 3 It is a structural schematic diagram of a personal non-repayment loan renewal review device based on a decision tree algorithm provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0052] The following is an explanation of some terminology concepts involved in the embodiments of the present application.

[0053] Refinancing without principal repayment: A financial product in which a customer applies to a bank for a loan of equal amount to repay the original loan principal before the original loan expires.

[0054] Decision tree algorithm: The decision tree algorithm is a supervised learning algorithm. Its basic principle is to split the data set through a series of questions, which are regarded as leaf nodes and internal nodes of the decision tree. Each branch of the decision tree represents a possible decision result, and each leaf node represents a final classification result. The process of making decisions using a decision tree is to start from the root node, test the corresponding feature attributes in the item to be classified, and select the output branch according to its value until the leaf node is reached, and the category stored in the leaf node is used as the decision result. In the embodiment of the present application, the decision tree algorithm is mainly used to classify according to the characteristics of multiple dimensions such as the known basic information of the customer, credit record, past cash flow information, etc. The decision tree contains thresholds for classifying the characteristics of each dimension. The output result of the decision tree algorithm is the customer's loan renewal evaluation result, and the customer's loan renewal evaluation result represents the possibility of whether the customer can renew the loan.

[0055] CART algorithm: Classification and regression tree algorithm, is an implementation of decision tree. CART algorithm is a binary recursive segmentation technique that divides the current sample into two sub-samples, so that each non-leaf node generated has two branches. Therefore, the decision tree generated by CART algorithm is a binary tree with a simple structure.

[0056] PCA algorithm: PCA algorithm, or principal component analysis algorithm, is the most widely used data dimensionality reduction algorithm. The PCA algorithm sequentially finds a set of mutually orthogonal coordinate axes from the original space and projects the data onto these coordinate axes to obtain a low-dimensional representation of the data.

[0057] The following is an example of an application scenario of the embodiment of the present application.

[0058] In the field of modern financial services, loan renewal business is crucial to maintaining liquidity and supporting the long-term development of customers. The traditional review method for personal loan renewal without repayment mainly relies on manual review by account managers. This process is not only time-consuming but also inefficient. Especially when processing a large number of loan renewal applications, it is easy to cause unstable review quality and service delays. With the development of Internet technology and big data, financial institutions have begun to explore more efficient loan renewal review methods to improve processing speed and decision-making accuracy.

[0059] In this context, the application of machine learning technology has gradually attracted attention. Machine learning, especially decision tree algorithms, has shown great potential in financial risk management and decision making due to its excellent classification ability and easy-to-interpret characteristics. By learning from historical data, decision trees can automatically identify the key factors that determine customer credit behavior, thereby making a quick and accurate assessment of future loan renewal applications.

[0060] However, the existing loan renewal review methods lack comprehensive evaluation and quality control of customer data. The customer's basic information, credit record, and past transaction information play a vital role in the loan renewal qualification assessment. These data not only reflect the customer's repayment ability, but also reveal their financial health and credit risk level. The existing technologies either fail to fully consider the customer's data, or even if the comprehensive data of the customer is collected, the data is not properly pre-processed to control the data quality.

[0061] In addition, the algorithm models used in existing technologies have poor interpretability and are difficult to promote and improve. Existing technologies currently use hidden Markov models or deep learning models. Although such models can achieve good classification results based on big data, they are difficult to interpret, which constitutes an obstacle to their application in fields such as finance that require high transparency and interpretability.

[0062] In summary, existing technologies lack comprehensive assessment and quality control of customer data. At the same time, the algorithm models used have poor interpretability and are difficult to promote and improve. This has largely hindered the application of artificial intelligence technology in the field of loan renewal qualification review.

[0063] In response to the above problems, the embodiment of the present application proposes a method for personal non-repayment loan renewal review based on a decision tree algorithm. The method obtains and pre-processes the comprehensive information of the customer, such as the customer's basic information, credit record, past transaction data, etc., and uses the decision tree algorithm for dimensionality reduction and model training to construct a decision model with good versatility. This can not only accurately assess the customer's loan renewal qualifications, but also effectively improve the work efficiency of the review process and speed up the loan renewal process, thereby partially or completely replacing the traditional manual review process. With this method, financial institutions can manage loan renewal business more accurately and efficiently, while reducing potential credit risks.

[0064] The following is an example of the method flow provided in the embodiments of the present application.

[0065] The loan renewal business process mainly includes loan renewal application, loan renewal approval and loan renewal signing.

[0066] In the loan renewal approval process, an approval algorithm will be introduced to check whether the customer meets the requirements, whether the loan elements need to be modified, and the circumstances of the modification.

[0067] Normally, the account manager needs to determine whether the current loan renewal application meets the conditions during the loan approval process. According to current business statistics, the vast majority of loan renewal application conditions are based on the previous loan made a year ago or even earlier. The basis for the loan, such as collateral, market environment, customer information, etc., has changed. It is necessary to comprehensively consider the various conditions of the previous loan, conduct quantitative analysis, and re-grant credit and pricing.

[0068] Please refer to the attached Figure 1 , attached Figure 1 The input parameters and output results of the decision tree algorithm in the embodiment of the present application are shown. Figure 2 , attached Figure 2 The flowchart of a method for reviewing personal non-repayment loan renewal based on a decision tree algorithm provided in an embodiment of the present application is shown. Figure 2 The method shown is applied to a computer device.

[0069] S210, obtaining a customer information data set. The customer information data set includes n-dimensional features of each of the m customers, and the n-dimensional features of each customer include the customer's basic information, the customer's credit record, and the customer's past transaction records.

[0070] The data acquired by the embodiment of the present application includes a customer information data set, and the customer information data set includes data such as the customer's basic information, the customer's credit record, and the customer's past cash flow information. If collateral is involved, the acquired data also includes collateral-related information. The customer's basic information includes age, gender, education level, job type, monthly income, etc. The customer's credit record includes credit information such as loan records and credit card usage records. The customer's past cash flow record includes historical tax data, monthly transaction volume, monthly transaction amount, etc. The collateral information includes the value of the collateral and the current status of the collateral market. In some embodiments, the customer information data set includes n-dimensional features of m customers. The customer information data set can be understood as an m*n matrix, and the n-dimensional features of each customer include the customer's basic information, the customer's credit record, and the customer's past cash flow information.

[0071] S220: Preprocess the customer information data set to obtain a preprocessed customer information data set.

[0072] After obtaining the data, in order to improve the data quality, the data needs to be preprocessed. Since there are many data sources and some data are manually entered, there are cases where values ​​are missing. For the above cases where values ​​are missing, the preprocessing process includes the following steps 1 to 4.

[0073] Step 1: Calculate the missing value ratio of all features in the customer information dataset. For example, for the features of each dimension in n dimensions, determine the missing value ratio of the dimension feature. The missing value ratio represents the ratio between the number of customers with missing dimension features among m customers and m. For example, for the ages of three customers, the customer information dataset contains the age of customer 1, which is 18 years old, and the customer information dataset does not contain the age of customer 2 and the age of customer 3. Therefore, the missing value ratio of the age dimension feature is 2 / 3.

[0074] Step 2: If the missing value ratio of a feature of a dimension exceeds a missing value ratio threshold, the feature of the dimension is deleted from the customer information data set. Exemplarily, the missing value ratio threshold is 30%.

[0075] Step 3: If the missing value ratio of the dimension feature does not exceed the missing value ratio threshold, the missing values ​​of the m customers in the dimension feature are filled with the average value of the m customers in the dimension feature. For example, the missing value ratio threshold is 30%.

[0076] Compared with the method of filling missing values ​​with fixed values, the method of filling with mean values ​​makes the filled data more consistent with the actual situation of the data, thereby improving the authenticity of the preprocessed data. In addition, it also avoids the problem of insufficient data volume caused by directly deleting features with missing values ​​accounting for no more than 30%, thereby increasing the amount of preprocessed data and thus improving the quality of the data.

[0077] Step 4: Normalize the features of each dimension in the customer information dataset after deleting features and filling missing values.

[0078] For example, for features of the same dimension, the mean and variance of the features of the dimension are calculated, and based on the mean and variance of the features of the dimension, the features of the dimension are converted to the range of (0, 1). By performing the normalization process, the impact of large differences in the values ​​of different features of the same dimension on the subsequent calculation accuracy can be reduced.

[0079] S230 , using a PCA algorithm to reduce the dimension of the preprocessed customer information data set to obtain a customer information data set after dimension reduction.

[0080] Since there are too many features and too few samples relative to the number of features, the generalization ability of the trained model is poor, so it is necessary to reduce the dimension of the features. This patent uses the PCA algorithm to obtain a low-dimensional representation of the data for the next step of the model training process.

[0081] The process of the PCA algorithm includes the following steps (1) to (4).

[0082] Step (1) normalizes the features of each dimension in the preprocessed customer information dataset to ensure that each feature has zero mean and unit variance. This is an important step because PCA is very sensitive to the scale of the data.

[0083] Specifically, for each dimension of the features of n dimensions in the preprocessed customer information data set, the average value of the feature of the dimension and the standard deviation of the feature of the dimension are calculated, and the feature of the dimension is standardized using the following formula.

[0084]

[0085] Among them, x is the original feature in the preprocessed customer information dataset, μ is the mean value of the feature, σ is the standard deviation of the feature, and z is the standardized result.

[0086] Step (2) determines the covariance matrix between the features of each dimension in the n-dimensional features.

[0087] After the customer information data set is standardized, the covariance matrix between the features of each dimension in the customer information data set is calculated. The covariance matrix is ​​used to characterize the correlation between features of different dimensions. For example, for the features of the i-th dimension among the features of n dimensions in the customer information data set, the covariance between the features of the i-th dimension and the features of the other (n-1) dimensions among the features of n dimensions except the i-th dimension and the covariance between the features of the i-th dimension and the features of the i-th dimension are determined to obtain the covariance matrix.

[0088] The covariance matrix is ​​defined as:

[0089]

[0090] Among them, Cov(X i , X j ) is feature X i and X j covariance, n is the number of samples, X ik and X jk They are feature X i and X j At the value of the kth sample, and It is feature X i and X j The sample mean of .

[0091] Step (3) performs eigendecomposition on the covariance matrix to obtain k eigenvalues ​​and k eigenvectors, where each eigenvalue corresponds to an eigenvector and the sum of the k eigenvalues ​​is 1.

[0092] Eigenvalues ​​and eigenvectors reveal the main direction of data change. Arrange the eigenvalues ​​from large to small, and the corresponding eigenvectors are the principal components of the data. The eigenvectors define a new feature space, and the eigenvalues ​​represent the data variance in the direction of each principal component.

[0093] Step (4) Sort the k eigenvalues ​​in descending order, and select the eigenvectors corresponding to the first p largest eigenvalues ​​from the k eigenvalues. The p eigenvectors are also called p principal components, so that the cumulative contribution rate reaches more than 85%, indicating that the p principal components reflect most of the information of the original data. The cumulative contribution rate is the sum of the first p eigenvalues. The p principal components are used as the input of the decision tree algorithm. The p principal components are the sum of the p eigenvalues ​​corresponding to the p eigenvectors.

[0094] S240. Based on the CART algorithm, a decision tree is trained using a target data set, where the target data set includes a dimensionally reduced customer information data set and a loan renewal evaluation result for each of the m customers.

[0095] The process of training a decision tree based on the CART algorithm is as follows:

[0096] (1) Select the target segmentation point.

[0097] The target split point can also be called the optimal split point. The split point is the threshold of a feature in a dimension within the decision tree. The decision tree determines the customer's loan renewal assessment result based on whether the customer's feature in this dimension is greater than the threshold. For example, if the decision tree determines that the customer is eligible for loan renewal based on the customer's age being greater than 20 years old and the monthly income being greater than 10,000 yuan, then 20 years old is the split point for the age dimension, and 10,000 yuan is the split point for the monthly income dimension.

[0098] Feature selection: For each dimension of the n-dimensional features and the loan renewal assessment results, all possible segmentation points of the features of the dimension are determined based on the CART algorithm to obtain multiple candidate segmentation points. The target segmentation point and the segmentation method corresponding to the target segmentation point are determined from the multiple candidate segmentation points.

[0099] Segmentation criteria: For each feature and segmentation point, the target dataset is divided into two subsets, and the impurity or variance after segmentation is calculated. The segmentation point that reduces the impurity or variance the most is selected to obtain the target segmentation point. The target dataset includes the customer information dataset after dimensionality reduction and the loan renewal evaluation results for each of the m customers.

[0100] The Gini index is used to evaluate the impurity of the segmentation. For example, the larger the Gini index, the higher the purity of the segmentation or the lower the impurity.

[0101] The Gini coefficient formula is:

[0102]

[0103] Among them, p i is the probability of class i in the split and j is the total number of classes.

[0104] (2) Recursively construct subtrees to obtain the initial decision tree.

[0105] Use the optimal split point to split the target dataset into two subsets.

[0106] Repeat steps (1) and (2) for each subset, recursively building a decision tree. Continue recursively until a stopping condition is met, such as the subset size is less than a threshold, the maximum tree depth is reached, or all records are of the same type.

[0107] (3) Prune the initial decision tree to obtain a pruned decision tree.

[0108] To prevent overfitting, post-pruning is performed. Each node of the decision tree is evaluated to determine whether its subtree should be pruned.

[0109] If pruning the subtree can improve the generalization ability of the model, pruning is performed. Specifically, for any node in the decision tree, the impurity of the segmentation after pruning the subtree of the node and the impurity of the segmentation of the subtree without pruning the node are determined. If the impurity of the segmentation after pruning the subtree of the node is reduced, the subtree of the node is pruned.

[0110] The complexity parameter α is used to balance the subtree complexity and the degree of fitting, and the optimal α value is found to maximize the generalization performance.

[0111] (4) Based on the test set, the hyperparameters in the pruned decision tree are adjusted to obtain the trained decision tree.

[0112] After obtaining a trained decision tree on the training set through the CART algorithm, the performance of the model is evaluated on the test set and the hyperparameter settings of the training model are continuously iterated until the model effect is optimal. After that, the decision tree model can be used for subsequent loan renewal qualification assessment.

[0113] S250: In response to receiving a loan renewal application request from a target customer, using a trained decision tree to determine a loan renewal evaluation result for the target customer based on information of the target customer.

[0114] Specifically, when a target customer applies for a loan renewal, the target customer's basic information, the target customer's credit record, and the target customer's past cash flow record are obtained, and the target customer's basic information, the target customer's credit record, and the target customer's past cash flow record are input into the trained decision tree, and the target customer's loan renewal evaluation result is output. The target customer's loan renewal evaluation result can provide a reference for business personnel to decide whether to approve the target customer's loan renewal application request, and the result feedback can be summarized to complete the loan renewal approval.

[0115] After the embodiment of the present application realizes manual approval through an algorithm, the approval speed is accelerated and the accuracy is improved. Furthermore, since the embodiment of the present application adopts a decision tree algorithm, it is more in line with the logic of deciding whether to allow a customer to renew the loan based on the customer's basic information, the customer's credit record, and the customer's past transaction record compared to the use of a hidden Markov model or a deep learning model. For example, the trained decision tree model contains a threshold A corresponding to the monthly income and a threshold B corresponding to the historical tax data. When the monthly income in the input customer information is greater than the threshold A and the historical tax data is greater than the threshold B, the loan renewal evaluation result output by the decision tree model is that the target customer can renew the loan, and the thresholds for classifying the input features of each dimension within the decision tree can be retrieved through the tool, so that it can explain why the decision tree model determines that the target customer can renew the loan based on the threshold A and the threshold B.

[0116] Figure 3 300 is a schematic diagram of a structure of a personal non-principal loan renewal review device based on a decision tree algorithm provided in an embodiment of the present application. The device 300 includes:

[0117] An acquisition unit 310 is used to acquire a customer information data set, where the customer information data set includes n-dimensional features of each of the m customers, where the n-dimensional features of each customer include the customer's basic information, the customer's credit record, and the customer's past transaction records;

[0118] A preprocessing unit 320, configured to preprocess the customer information data set to obtain a preprocessed customer information data set;

[0119] A dimension reduction unit 330 is used to reduce the dimension of the preprocessed customer information data set by using a PCA algorithm to obtain a dimension-reduced customer information data set;

[0120] A training unit 340 is used to train a decision tree using a target data set based on a CART algorithm, where the target data set includes a dimensionally reduced customer information data set and a loan renewal evaluation result for each of the m customers;

[0121] The determination unit 350 is used to determine the loan renewal evaluation result of the target customer based on the target customer's information using the trained decision tree in response to receiving the target customer's loan renewal application request, where the target customer's information includes n-dimensional features of the target customer.

[0122] In some embodiments, the preprocessing unit 320 is used to determine the missing value ratio of the features of each dimension in the n dimensions, where the missing value ratio represents the ratio between the number of customers who are missing the features of the dimension among the m customers in the customer information dataset and m; if the missing value ratio of the features of the dimension exceeds the missing value ratio threshold, the features of the dimension are deleted from the customer information dataset; if the missing value ratio of the features of the dimension does not exceed the missing value ratio threshold, the missing values ​​of the features of the m customers in the dimension are filled with the average value of the features of the m customers in the dimension; and the features of each dimension in the customer information dataset after the features are deleted and the missing values ​​are filled are normalized.

[0123] In some embodiments, the dimension reduction unit 330 is used to standardize the features of each dimension in the preprocessed customer information data set based on the average value of the features of the dimension and the standard deviation of the features of the dimension to obtain standardized features; determine the covariance matrix between the features of each dimension in the features of n dimensions; decompose the covariance matrix to obtain eigenvalues ​​and eigenvectors; select eigenvectors corresponding to the first p eigenvalues ​​from the eigenvalues, and the sum of the p eigenvalues ​​is greater than the contribution rate threshold, and use the selected p eigenvectors as the customer information data set after dimensionality reduction.

[0124] In some embodiments, the training unit 340 is used to determine the target segmentation point of each dimension of the features of n dimensions based on the target data set; recursively construct a subtree based on the target segmentation point to obtain an initial decision tree; prune the initial decision tree to obtain a pruned decision tree; and adjust the hyperparameters in the pruned decision tree based on the test set to obtain a trained decision tree.

[0125] In some embodiments, the training unit 340 is used to determine multiple candidate segmentation points for the features of each dimension of the n-dimensional features and the renewal loan evaluation results; based on the candidate segmentation points, the target data set is divided into two subsets to determine the impurity or variance after segmentation; and the segmentation point that reduces the impurity or variance the most is selected to obtain the target segmentation point.

[0126] In some embodiments, the training unit 340 is used to determine, for each node in the initial decision tree, the impurity of the segmentation after pruning the node's subtree and the impurity of the segmentation of the node's subtree without pruning it; if the impurity of the segmentation after pruning the node's subtree is reduced, the node's subtree is pruned from the initial decision tree.

[0127] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0128] A refers to B, which means that A is the same as B or A is a simple variant of B.

[0129] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loading and executing computer program instructions on a computer, the process or function described in accordance with the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrations. Available media can be magnetic media, (e.g., floppy disk, hard disk, tape), optical media (e.g., DVD), or semiconductor media (e.g., solid-state hard disk Solid State Disk (SSD)), etc.

[0130] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for reviewing personal principal-free loan renewal based on a decision tree algorithm, characterized in that: The method comprises: Obtain a customer information data set, wherein the customer information data set includes n-dimensional features of each customer among m customers, and the n-dimensional features of each customer include basic information of the customer, the credit record of the customer, and the past transaction record of the customer; Preprocessing the customer information data set to obtain a preprocessed customer information data set; Using the PCA algorithm to reduce the dimension of the preprocessed customer information data set to obtain a customer information data set after dimension reduction; Based on the CART algorithm, a decision tree is trained using a target data set, wherein the target data set includes the customer information data set after dimensionality reduction and a loan renewal evaluation result for each of the m customers; In response to receiving a loan renewal application request from a target customer, a loan renewal evaluation result of the target customer is determined based on information of the target customer using a trained decision tree, wherein the information of the target customer includes features of the n dimensions of the target customer.

2. The method according to claim 1, characterized in that The preprocessing of the customer information data set to obtain a preprocessed customer information data set includes: For the features of each dimension in the n dimensions, determine the missing value ratio of the features of the dimension, where the missing value ratio represents the ratio between the number of customers who lack the features of the dimension among the m customers in the customer information dataset and m; If the missing value ratio of the feature of the dimension exceeds the missing value ratio threshold, delete the feature of the dimension from the customer information data set; if the missing value ratio of the feature of the dimension does not exceed the missing value ratio threshold, use the average value of the feature of the dimension of m customers to fill the missing values ​​of the feature of the dimension of m customers; The features of each dimension in the customer information dataset are normalized after deleting features and filling missing values.

3. The method according to claim 1, characterized in that The PCA algorithm is used to reduce the dimension of the preprocessed customer information data set to obtain a customer information data set after dimension reduction, including: For the feature of each dimension in the preprocessed customer information data set, based on the average value of the feature of the dimension and the standard deviation of the feature of the dimension, the feature of the dimension is standardized to obtain a standardized feature; Determine the covariance matrix between the features of each dimension in the features of the n dimensions; Decomposing the covariance matrix to obtain eigenvalues ​​and eigenvectors; Select eigenvectors corresponding to the first p eigenvalues ​​from the eigenvalues, the sum of the p eigenvalues ​​being greater than a contribution rate threshold, and use the selected p eigenvectors as a customer information data set after dimensionality reduction.

4. The method according to claim 1, characterized in that: The method of training a decision tree using a target data set based on the CART algorithm includes: Based on the target data set, determine a target segmentation point of a feature of each dimension of the features of the n dimensions; Recursively constructing a subtree based on the target segmentation point to obtain an initial decision tree; Pruning the initial decision tree to obtain a pruned decision tree; The hyperparameters in the pruned decision tree are adjusted based on the test set to obtain the trained decision tree.

5. The method according to claim 4, characterized in that The step of determining a target segmentation point of a feature of each dimension of the features of the n dimensions based on the target data set includes: For each dimension of the n-dimensional features and the loan renewal assessment result, multiple candidate split points are determined; Based on the candidate segmentation points, the target data set is divided into two subsets, and the impurity or variance after segmentation is determined; Select the split point that reduces the most impurity or variance to get the target split point.

6. The method according to claim 4, characterized in that The pruning of the initial decision tree to obtain a pruned decision tree includes: For each node in the initial decision tree, determining the impurity of the segmentation after pruning the subtree of the node and the impurity of the segmentation of the subtree without pruning the node; If the impurity of the segmentation is reduced after pruning the subtree of the node, the subtree of the node is pruned from the initial decision tree.

7. A personal non-repayment loan renewal review device based on a decision tree algorithm, characterized in that: The device comprises: An acquisition unit is used to acquire a customer information data set, wherein the customer information data set includes n-dimensional features of each customer among m customers, and the n-dimensional features of each customer include basic information of the customer, the credit record of the customer, and the past transaction record of the customer; A preprocessing unit, used to preprocess the customer information data set to obtain a preprocessed customer information data set; A dimension reduction unit, used for reducing the dimension of the preprocessed customer information data set by using a PCA algorithm to obtain a dimension-reduced customer information data set; A training unit, configured to train a decision tree using a target data set based on a CART algorithm, wherein the target data set includes the customer information data set after dimensionality reduction and a loan renewal evaluation result for each of the m customers; A determination unit is used to, in response to receiving a loan renewal application request from a target customer, determine a loan renewal evaluation result for the target customer based on information of the target customer using a trained decision tree, wherein the information of the target customer includes features of the target customer in the n dimensions.

8. The device according to claim 7, characterized in that The preprocessing unit is used to determine the missing value ratio of the features of each dimension in the n dimensions, wherein the missing value ratio represents the ratio between the number of customers who are missing the features of the dimension among the m customers in the customer information dataset and m; if the missing value ratio of the features of the dimension exceeds a missing value ratio threshold, delete the features of the dimension from the customer information dataset; if the missing value ratio of the features of the dimension does not exceed the missing value ratio threshold, fill the missing values ​​of the features of the m customers in the dimension with the average value of the features of the m customers in the dimension; and normalize the features of each dimension in the customer information dataset after deleting the features and filling the missing values.

9. The device according to claim 7, characterized in that The dimension reduction unit is used to standardize the features of each dimension in the preprocessed customer information data set based on the average value of the features of the dimension and the standard deviation of the features of the dimension to obtain the standardized features; determine the covariance matrix between the features of each dimension in the n dimensional features; decompose the covariance matrix to obtain eigenvalues ​​and eigenvectors; select eigenvectors corresponding to the first p eigenvalues ​​from the eigenvalues, the sum of the p eigenvalues ​​being greater than a contribution rate threshold, and use the selected p eigenvectors as the customer information data set after dimensionality reduction.

10. The device according to claim 7, characterized in that The training unit is used to determine a target segmentation point of a feature of each dimension of the features of the n dimensions based on the target data set; Recursively constructing a subtree based on the target segmentation point to obtain an initial decision tree; The initial decision tree is pruned to obtain a pruned decision tree; and hyperparameters in the pruned decision tree are adjusted based on a test set to obtain a trained decision tree.

Images