Financial customer credit classification method and device, electronic equipment and program product

Abstract

The invention discloses a financial customer credit classification method and device, electronic equipment and a program product, and relates to the financial science and technology field, and the method comprises the steps: employing a decision classifier to carry out the classification screening of a target data set and a to-be-predicted credit data set, and carrying out the parameter adjustment of a multi-classification machine learning model through a final data set, a multi-classification machine learning model is adopted to execute classification prediction of financial customer credit, the credit category of the financial customer is output, and the target data set is a data set with balanced data distribution obtained after the initial balance data set and the synthetic data set are merged. The initial balance data set is generated by performing adaptive synthesis and oversampling on minority customer samples in the original data set, and the synthesized data set is generated by performing deep learning on the initial balance data set by using a generative adversarial network model. According to the invention, the technical problem that the classification accuracy is affected due to low classification precision of minority types of customers during financial customer credit classification in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of financial technology or other related fields, and more specifically, to a method and apparatus for classifying the credit of financial customers, as well as electronic devices and program products. Background Technology

[0002] In financial institutions, accurately assessing a customer's credit category is crucial for risk management, loan approval, and personalized services. However, due to the uneven distribution of customer numbers across different categories in the dataset, especially when the number of samples in the minority category (such as high-risk customers) is far less than that in the majority category (such as low-risk customers), traditional classification models tend to favor the majority category during training, leading to a decrease in the classification accuracy of minority customers.

[0003] In the multi-class classification scenario of bank customer credit, the global distribution between categories was not fully considered when generating synthetic samples, which may result in the generated samples failing to represent the diversity of the real dataset. This is especially true in multi-class classification problems, where the boundaries and distributions between categories are more complex.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This invention provides a method, apparatus, electronic device, and program product for classifying the credit of financial customers, in order to at least solve the technical problem in the related art where the classification accuracy of a few types of customers is low, affecting the accuracy of classification.

[0006] To achieve the above objectives, according to one aspect of this application, a method for classifying the credit of financial customers is provided, comprising: receiving basic information of financial customers and a credit dataset to be predicted; classifying and filtering the target dataset and the credit dataset to be predicted using a preset decision classifier to form a final dataset; adjusting the parameters of a multi-class machine learning model using the final dataset, wherein the target dataset is a dataset with balanced data distribution obtained by merging an initial balanced dataset and a synthetic dataset, the initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset, and the synthetic dataset is a dataset generated by performing deep learning on the initial balanced dataset using a preset generative adversarial network model, and the initial balanced dataset and the synthetic dataset have the same number of data; and performing classification prediction of the credit of financial customers using the adjusted multi-class machine learning model to output the credit category of the financial customer.

[0007] Optionally, generating the initial balanced dataset includes: preprocessing the obtained original dataset of financial institution customers, wherein each piece of data in the original dataset includes at least: basic customer information, customer feature vector, and corresponding data label, wherein the data label includes at least: customer credit category; inputting the preprocessed original dataset into an adaptive synthetic oversampling model, wherein the adaptive synthetic oversampling model calculates K nearest neighbor samples for minority class samples in the original dataset, wherein the minority class samples refer to samples in the original dataset whose data volume corresponding to the customer credit category is less than a preset data volume threshold, and K is a positive integer greater than 1; based on the K nearest neighbor samples, calculating the number of synthetic samples to be generated for each minority class sample in the original dataset, and interpolating between the minority class samples and the minority class nearest neighbor samples to generate the number of synthetic samples to be generated; and combining the synthetic samples corresponding to all the minority class samples to generate the initial balanced dataset.

[0008] Optionally, the step of preprocessing the obtained raw dataset of financial institution customers includes: cleaning the obtained raw dataset of financial institution customers, wherein the data cleaning method includes at least one of the following: missing value imputation and outlier removal; and standardizing the numerical features in the cleaned raw dataset.

[0009] Optionally, the step of calculating the number of minority class samples to be generated in the original dataset based on the K nearest neighbor samples includes: calculating the proportion of majority class samples in the total samples among the K nearest neighbor samples of each minority class sample, and determining the difficulty coefficient corresponding to the minority class sample based on the sample proportion; for each minority class sample, calculating the total number of synthetic samples to be generated based on the total number of minority class samples and the number of majority class samples, and allocating the number of minority class samples to be generated based on the difficulty coefficient and the total number of synthetic samples.

[0010] Optionally, the step of interpolating between the minority class sample and the minority class nearest neighbor sample to generate the synthetic sample corresponding to the number to be generated includes: randomly selecting the minority class sample and the minority class nearest neighbor sample of the minority class sample; generating a random number corresponding to the minority class sample and the minority class nearest neighbor sample; and performing linear interpolation between the minority class sample and the minority class nearest neighbor sample based on the minority class sample, the minority class nearest neighbor sample and the random number to generate the synthetic sample corresponding to the number to be generated.

[0011] Optionally, the step of classifying and filtering the target dataset and the credit dataset to be predicted using a preset decision classifier to form the final dataset includes: removing the data labels from each data sample in the target dataset and setting the data labels to a first value; setting the data labels of each customer data sample in the credit dataset to be predicted to a second value, wherein the second value is different from the first value; using the target dataset and the credit dataset to be predicted with the labels set, the preset decision classifier outputs the label prediction probability corresponding to each data sample in the target dataset; and filtering out data samples whose label prediction probabilities are greater than a preset probability threshold to obtain the final dataset.

[0012] Optionally, when the multi-class machine learning model is a multilayer perceptron model, it includes: a neural network architecture with two hidden layers, the first hidden layer including a first set of neurons, the second hidden layer including a second set of neurons, all hidden layers using the ReLU activation function, wherein the first set of neurons is greater than the second set of neurons; an output layer using the Softmax activation function; wherein, weighted cross-entropy is used as the loss function during the training of the multi-class machine learning model.

[0013] According to another aspect of the present invention, a financial customer credit classification apparatus is also provided, comprising: a dataset receiving unit for receiving basic information of financial customers and a credit dataset to be predicted; a data classification and filtering unit for classifying and filtering the target dataset and the credit dataset to be predicted using a preset decision classifier to form a final dataset, and adjusting the parameters of a multi-class machine learning model using the final dataset, wherein the target dataset is a dataset with balanced data distribution obtained by merging an initial balanced dataset and a synthetic dataset, the initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset, and the synthetic dataset is a dataset generated by performing deep learning on the initial balanced dataset using a preset generative adversarial network model, and the initial balanced dataset and the synthetic dataset have the same number of data; and a credit classification unit for performing financial customer credit classification prediction using the adjusted multi-class machine learning model and outputting the credit category of the financial customer.

[0014] Optionally, the financial customer credit classification device further includes, when generating the initial balanced dataset: a data preprocessing unit, used to preprocess the acquired original dataset of financial institution customers, wherein each piece of data in the original dataset includes at least: basic customer information, customer feature vector, and corresponding data label, wherein the data label includes at least: customer credit category; a nearest neighbor sample calculation unit, used to input the preprocessed original dataset into an adaptive synthetic oversampling model, wherein the adaptive synthetic oversampling model calculates K nearest neighbor samples for the minority class samples in the original dataset, wherein the minority class samples refer to samples in the original dataset whose data volume corresponding to the customer credit category is less than a preset data volume threshold, and K is a positive integer greater than 1; a generation quantity calculation unit, used to calculate the number of to-be-generated samples for each minority class sample in the original dataset based on the K nearest neighbor samples, and interpolate between the minority class samples and the minority class nearest neighbor samples to generate the number of synthetic samples to be generated; and a balanced dataset generation unit, used to synthesize the synthetic samples corresponding to all the minority class samples to generate the initial balanced dataset.

[0015] Optionally, the data preprocessing unit includes: a data cleaning module for cleaning the acquired raw dataset of financial institution customers, wherein the data cleaning method includes at least one of the following: missing value imputation and outlier removal; and a standardization module for standardizing the numerical features in the cleaned raw dataset.

[0016] Optionally, the generation quantity calculation unit includes: a difficulty coefficient calculation module, used to calculate the proportion of majority class samples in the K nearest neighbor samples of each minority class sample to the total samples, and determine the difficulty coefficient corresponding to the minority class sample based on the sample proportion; and a generation quantity allocation module, used to calculate the total number of synthetic samples to be generated for each minority class sample based on the total number of minority class samples and the number of majority class samples, and allocate the generation quantity of each minority class sample based on the difficulty coefficient and the total number of synthetic samples.

[0017] Optionally, when the quantity calculation unit generates a synthetic sample corresponding to the quantity to be generated by interpolating between the minority class sample and the minority class nearest neighbor sample, it includes: a random selection module for randomly selecting one of the minority class samples and its minority class nearest neighbor sample; a random number generation module for generating a random number corresponding to the minority class sample and its minority class nearest neighbor sample; and a synthetic sample generation module for performing linear interpolation between the minority class sample and its minority class nearest neighbor sample based on the minority class sample, the minority class nearest neighbor sample, and the random number to generate the synthetic sample of the quantity to be generated.

[0018] Optionally, the data classification and filtering unit includes: a first label setting module, used to remove the data labels from each data sample in the target dataset and set the data labels to a first value; a second label setting module, used to set the data labels of each customer data sample in the credit dataset to be predicted to a second value, wherein the second value is different from the first value; a label prediction probability output module, used to output the label prediction probability corresponding to each data sample in the target dataset by the preset decision classifier using the target dataset with the labels set and the credit dataset to be predicted; and a dataset classification and filtering module, used to filter out data samples with label prediction probabilities greater than a preset probability threshold to obtain the final dataset.

[0019] Optionally, when the multi-class machine learning model is a multilayer perceptron model, it includes: a neural network architecture with two hidden layers, the first hidden layer including a first set of neurons, the second hidden layer including a second set of neurons, all hidden layers using the ReLU activation function, wherein the first set of neurons is greater than the second set of neurons; an output layer using the Softmax activation function; wherein, weighted cross-entropy is used as the loss function during the training of the multi-class machine learning model.

[0020] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the financial customer credit classification method described above.

[0021] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the financial customer credit classification method described above.

[0022] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the financial customer credit classification method described in any one of the above embodiments.

[0023] In this disclosure, basic information of financial customers and a credit dataset to be predicted are received; a pre-set decision classifier is used to classify and filter the target dataset and the credit dataset to be predicted to form a final dataset; the parameters of a multi-class machine learning model are adjusted using the final dataset. The target dataset is a dataset with balanced data distribution obtained by merging an initial balanced dataset and a synthetic dataset. The initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset. The synthetic dataset is a dataset generated by deep learning the initial balanced dataset using a pre-set generative adversarial network model. The initial balanced dataset and the synthetic dataset have the same number of data points. The adjusted multi-class machine learning model is used to perform classification prediction of financial customer credit and output the credit category of the financial customer.

[0024] Based on the aforementioned disclosures, an adaptive synthetic oversampling strategy can be used to adaptively increase the representativeness of minority class samples, alleviate data imbalance, and enable multi-class machine learning models to learn features of all classes fairly. High-quality synthetic datasets are generated through deep learning via generative adversarial networks, which not only increases the number of samples but also enhances data diversity, ensuring the comprehensiveness and complexity of the data during model training. Using a pre-defined decision classifier for data filtering ensures that the distribution of the generated synthetic data is similar to that of the actual predicted data, thereby improving the generalization ability of the final multi-class machine learning model and enabling it to perform better when facing new data. This significantly improves the classification accuracy of minority class customers and enhances the stability and accuracy of financial customer credit classification, thus solving the technical problem of low classification accuracy for minority class customers in financial customer credit classification, which affects classification accuracy. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0026] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a classification method for financial customer credit is shown.

[0027] Figure 2 This is a flowchart of an optional method for classifying the credit of financial customers according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of an optional financial customer credit classification method according to an embodiment of the present invention;

[0029] Figure 4This is a schematic diagram of an optional financial customer credit classification device according to an embodiment of the present invention;

[0030] Figure 5 This is a structural block diagram of an electronic device that performs a method for classifying the credit of financial customers according to an embodiment of this application. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0033] To facilitate understanding of the present invention by those skilled in the art, some terms or nouns involved in the various embodiments of the present invention are explained below:

[0034] Synthetic Minority Over-sampling Technique (SMOTE) is a data preprocessing technique used to address class imbalance. It increases the number of minority class samples by generating synthetic samples (i.e., interpolation) in the feature space of minority class samples, thereby balancing the proportion of samples from each class in the dataset.

[0035] Adaptive Synthetic Sampling (ADASYN) is an improved version of SMOTE. It not only generates synthetic samples but also pays special attention to minority class samples that are more difficult to classify. By adaptively determining the number of synthetic samples for each minority class sample, ADASYN attempts to improve the model's ability to identify boundary samples, thereby enhancing the overall performance of the classification model.

[0036] Conditional Generative Adversarial Networks (CTGANs) are a variant of Generative Adversarial Networks (GANs) specifically designed for generating high-quality tabular data. They consist of two main components: a generator and a discriminator. The generator learns the distribution of real data to produce new samples similar to the original data, while the discriminator attempts to distinguish between the generated samples and real samples. Through this competitive optimization process, CTGANs can generate diverse new samples that conform to the data distribution, making them particularly suitable for handling class imbalance problems and enhancing the diversity and representativeness of datasets.

[0037] A Multilayer Perceptron (MLP) is a feedforward neural network composed of multiple layers of neurons, including an input layer, one or more hidden layers, and an output layer. It performs classification or regression tasks by learning complex nonlinear relationships in the input data. In this invention, an MLP is used as the final classifier to learn a customer credit classification model from a preprocessed and augmented dataset and make predictions.

[0038] Category Boosting (CatBoost) is a machine learning algorithm. In this invention, CatBoost is used to filter the generated synthetic samples to ensure that these samples are similar to the distribution of the target data to be predicted, thereby improving the accuracy of the final credit category prediction model.

[0039] It should be noted that the financial customer credit classification method and apparatus in this disclosure can be used in the fintech field to classify financial customer credit based on an adaptive oversampling strategy and a generative adversarial network, and can also be used in any field other than fintech to classify financial customer credit based on an adaptive oversampling strategy and a generative adversarial network. This disclosure does not limit the application field of the financial customer credit classification method and apparatus.

[0040] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected in this public disclosure are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or organization.

[0041] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0042] The following embodiments of the present invention can be applied to various systems / applications / devices for classifying the credit of financial customers. The present invention is applicable to customer credit assessment in the financial field, particularly in bank customer credit classification prediction. For example, when reviewing loan applications, banks need to accurately assess the applicant's credit category. The present invention improves the accuracy of the classification model by enhancing data balance and diversity, thereby helping banks to more accurately decide whether to approve loan applications and reduce credit risk. Based on the customer's credit category, banks can also provide personalized services and products, such as customized loan programs and priority services. The present invention improves the accuracy of credit category prediction, enabling banks to more accurately identify customers with different credit categories, thereby optimizing the customer service experience.

[0043] This invention employs an adaptive oversampling method to adaptively increase the representativeness of minority class samples, effectively alleviating the data imbalance problem and enabling the model to learn features from all classes fairly during training. High-quality synthetic data is generated through generative adversarial networks, which not only increases the number of samples but also enhances data diversity, ensuring the comprehensiveness and complexity of the data during model training. A classifier is used for data filtering, ensuring that the distribution of the generated synthetic data is similar to that of the actual prediction data, thereby improving the generalization ability of the final multi-class machine learning model and enabling it to perform better when faced with new data.

[0044] The present invention will now be described in detail with reference to various embodiments.

[0045] Example 1

[0046] According to an embodiment of the present invention, an embodiment of a method for classifying the credit of financial customers is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0047] The financial customer credit classification method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method for classifying the credit of financial customers is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 The processor 102 (illustrated as 102a, 102b, ..., 102n) may include, but is not limited to, a microprocessor MCU (Microcontroller Unit) or a programmable gate array (FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0048] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0049] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the financial customer credit classification method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned financial customer credit classification method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0050] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0051] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0052] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for classifying the credit of financial clients is shown. Figure 2 This is a flowchart of an optional financial customer credit classification method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0053] Step S201: Receive basic information of financial customers and the credit dataset to be predicted.

[0054] In this embodiment, the basic information may include multi-dimensional data such as the customer's identity, financial status, transaction records, and credit history. The credit dataset to be predicted consists of data of new customers who are about to be rated. This data is unprocessed and contains the same type of characteristics as the new customers.

[0055] Step S202: A preset decision classifier is used to classify and filter the target dataset and the credit dataset to be predicted to form the final dataset. The parameters of the multi-class machine learning model are adjusted using the final dataset. The target dataset is a dataset with balanced data distribution obtained by merging the initial balanced dataset and the synthetic dataset. The initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset. The synthetic dataset is a dataset generated by using a preset generative adversarial network model to perform deep learning on the initial balanced dataset. The initial balanced dataset and the synthetic dataset have the same number of data.

[0056] Optionally, the process of generating the initial balanced dataset includes: preprocessing the obtained original dataset of financial institution customers, wherein each data point in the original dataset includes at least: basic customer information, customer feature vector, and corresponding data label, and the data label includes at least: customer credit category; inputting the preprocessed original dataset into an adaptive synthetic oversampling model, which calculates K nearest neighbor samples for the minority class samples in the original dataset, wherein the minority class samples refer to samples in the original dataset whose data volume corresponding to the customer credit category is less than a preset data volume threshold, and K is a positive integer greater than 1; based on the K nearest neighbor samples, calculating the number of synthetic samples to be generated for each minority class sample in the original dataset, and interpolating between the minority class samples and the minority class nearest neighbor samples to generate the number of synthetic samples to be generated; and combining the synthetic samples corresponding to all minority class samples to generate the initial balanced dataset.

[0057] In the initial stage, this embodiment preprocesses the raw customer data collected by financial institutions. Each data point in the raw dataset includes at least the customer's basic information, feature vector, and data label, where the data label explicitly indicates the customer's credit category. Optionally, the preprocessing steps for the obtained raw dataset of financial institution customers include: data cleaning, wherein the data cleaning methods include at least one of the following: missing value imputation and outlier removal; and standardization of the numerical features in the cleaned raw dataset. The preprocessing process involves data cleaning, standardization, and encoding to ensure that all data is suitable for input into an adaptive synthetic oversampling model (such as ADASYN), effectively removing outliers and handling missing data, while converting categorical features into numerical or vector forms that machine learning algorithms can understand.

[0058] Furthermore, the preprocessed dataset can be input into an adaptive synthetic oversampling model. This model first identifies minority class samples in the original dataset, namely customers whose credit classification data volume is lower than a preset data volume threshold. The model calculates the K nearest neighbor samples for each minority class sample, where K is a predefined positive integer, usually set to a value greater than 1, in order to fully consider the neighborhood environment of the minority class samples (e.g., setting K to 5 or 7). The calculation process uses distance metrics, such as Euclidean distance, to determine which samples are closest to the current minority class sample.

[0059] Subsequently, based on the calculated K nearest neighbor samples, an adaptive synthetic oversampling model can be used to further calculate the number of synthetic samples to be generated for each minority class sample. The calculation of the number of synthetic samples is based on the difficulty coefficient of the minority class samples, which can selectively generate more samples located near the decision boundary. These samples are often the most difficult to classify, so generating synthetic samples can significantly improve the classifier's ability to recognize these samples. Optionally, the step of calculating the number of synthetic samples to be generated for each minority class sample in the original dataset based on the K nearest neighbor samples includes: calculating the proportion of majority class samples in the total samples among the K nearest neighbor samples of each minority class sample, and determining the difficulty coefficient corresponding to the minority class sample based on the sample proportion; for each minority class sample, calculating the total number of synthetic samples to be generated based on the total number of minority class samples and the number of majority class samples, and allocating the number of synthetic samples to be generated for each minority class sample based on the difficulty coefficient and the total number of synthetic samples.

[0060] This embodiment first calculates the proportion of majority class samples among the K nearest neighbors of each minority class sample. This proportion reflects the "isolation" of the minority class sample in the dataset, i.e., the density of the surrounding majority class samples. Based on this sample proportion, a difficulty coefficient corresponding to the minority class sample is determined. The higher the difficulty coefficient, the more majority class samples are around the sample, thus making it more difficult for the model to learn from that sample, requiring the generation of more synthetic samples to enhance its representativeness. For each minority class sample, this embodiment calculates the total number of synthetic samples to be generated based on the current total number of minority class samples and the number of majority class samples. The goal is to balance the dataset, i.e., to make the number of minority class samples close to or equal to the number of majority class samples by generating synthetic samples. Based on the difficulty coefficient and the total number of synthetic samples, this embodiment allocates the number of samples to be generated for each minority class sample, ensuring that more difficult samples receive more synthetic sample support, thereby achieving a better representation in the dataset.

[0061] Optionally, the step of generating a synthetic sample corresponding to the number to be generated by performing linear interpolation between the minority class sample and the minority class nearest neighbor sample includes: randomly selecting a minority class sample and the minority class nearest neighbor sample of the minority class sample; generating a random number corresponding to the minority class sample and the minority class nearest neighbor sample; and generating a synthetic sample corresponding to the number to be generated by performing linear interpolation between the minority class sample and the minority class nearest neighbor sample based on the minority class sample, the minority class nearest neighbor sample and the random number.

[0062] Using a linear interpolation method, this embodiment generates a specific number of synthetic samples between minority class samples and their K nearest neighbors. For each pair of samples, the model selects a random point in their feature space and then generates an interpolation path from the minority class sample to its nearest neighbor based on the position of that point, thereby creating a new synthetic sample. Finally, this embodiment combines all the generated synthetic samples and merges them with the remaining samples in the original dataset to form an initial balanced dataset. The dataset's data volume for each credit category is enhanced, especially for the minority class samples, significantly improving the data imbalance problem.

[0063] Optionally, the step of classifying and filtering the target dataset and the credit dataset to be predicted using a preset decision classifier to form the final dataset includes: removing the data labels from each data sample in the target dataset and setting the data labels to a first value; setting the data labels to a second value for each customer data sample in the credit dataset to be predicted, wherein the second value is different from the first value; using the labeled target dataset and the credit dataset to be predicted, the preset decision classifier outputs the label prediction probability corresponding to each data sample in the target dataset; and filtering out data samples whose label prediction probabilities are greater than a preset probability threshold to obtain the final dataset.

[0064] In this embodiment, for each data sample in the target dataset, the original credit category data label can be removed and uniformly set to a new value, referred to as the first value, such as 0, thereby converting the target dataset into one of the categories in a binary classification problem. Simultaneously, the data label for each customer data sample in the credit dataset to be predicted is uniformly set to a second value, such as 1, ensuring a difference in numerical representation from the first value, thus labeling the dataset to be predicted as the other category in the binary classification problem. Using the pre-labeled target dataset and the credit dataset to be predicted, this embodiment uses a preset decision classifier (such as a CatBoost classifier) ​​for training and prediction. After training, the classifier can output the predicted label probability for each data sample in the target dataset, i.e., the probability value of belonging to the second value (the category of the dataset to be predicted). The predicted probability reflects the similarity between the distribution of the synthetic sample and the dataset to be predicted.

[0065] In some embodiments, this embodiment sets a predefined probability threshold to filter out data samples in the target dataset whose label prediction probability is greater than the threshold. The probability threshold can be set based on business needs and data characteristics. For example, it can be set to 80%. Only synthetic samples that are considered by the classifier to be highly similar to the distribution of the dataset to be predicted, i.e., whose probability value exceeds 80%, will be included in the final dataset. This ensures that the final dataset not only matches the dataset to be predicted in quantity, but also that the sample distribution is closer to real customer data in terms of quality, thereby improving the model's generalization ability and prediction accuracy.

[0066] Step S203: Use the adjusted multi-class machine learning model to perform classification prediction of financial customer credit and output the credit category of the financial customer.

[0067] It should be noted that the multi-class machine learning model in this embodiment can be an MLP (Multilayer Perceptron) model, or other multi-class machine learning algorithms, such as logistic regression, support vector machine, decision tree, random forest, gradient boosting tree, k-nearest neighbors, artificial neural network, etc.

[0068] Optionally, when the multi-class machine learning model is a multilayer perceptron model, it includes: a neural network architecture with two hidden layers, the first hidden layer including a first set of neurons, the second hidden layer including a second set of neurons, all hidden layers using the ReLU activation function, wherein the number of neurons in the first set is greater than the number of neurons in the second set; an output layer using the Softmax activation function; wherein, weighted cross-entropy is used as the loss function during the training of the multi-class machine learning model.

[0069] In this embodiment, the multi-class machine learning model adopts a two-layer hidden layer structure. The first hidden layer has a large number of neurons, which enhances the model's ability to extract features. The second hidden layer contains a smaller number of neurons, which aims to reduce complexity and promote feature integration. All hidden layers use ReLU (Rectified Linear Unit) as the activation function. ReLU can effectively alleviate the gradient vanishing problem, accelerate model convergence, and maintain sufficient non-linear expressive power, making it particularly suitable for processing the complex features of financial data.

[0070] It should be noted that the output layer in this embodiment directly maps the model's predictions to different credit categories for financial customers. A Softmax activation function is used to convert the neural network output into a probability distribution, ensuring that the sum of the probabilities of all predicted categories is 1, thereby assigning each customer a most likely credit category. Furthermore, to address the challenges posed by data imbalance during the training of the multi-class machine learning model, this embodiment employs weighted cross-entropy as the loss function. Weighted cross-entropy assigns higher weights to minority class samples, ensuring that during training, the model not only prioritizes the prediction of the majority class but also fully learns the features of the minority class. The use of this loss function helps improve the model's prediction accuracy and stability for all categories, especially the minority class.

[0071] Once the model is trained and the parameters are adjusted to their optimal state, this embodiment uses the adjusted multi-class machine learning model to perform classification prediction of financial customer credit. The model receives input from the final dataset (which consists of high-quality original samples and filtered synthetic samples, optimized through multiple steps such as preprocessing, data balancing, and deep synthesis). Based on this dataset, the model can more accurately identify the credit risk characteristics of customers and output the corresponding credit category.

[0072] Through the above steps, basic information of financial customers and a credit dataset to be predicted can be received. A preset decision classifier is used to classify and filter the target dataset and the credit dataset to be predicted, forming a final dataset. The parameters of the multi-class machine learning model are adjusted using the final dataset. The target dataset is a dataset with balanced data distribution obtained by merging the initial balanced dataset and the synthetic dataset. The initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset. The synthetic dataset is a dataset generated by using a preset generative adversarial network model to perform deep learning on the initial balanced dataset. The initial balanced dataset and the synthetic dataset have the same number of data. The adjusted multi-class machine learning model is used to perform classification prediction of financial customer credit and output the credit category of the financial customer. In this embodiment, an adaptive synthetic oversampling strategy can be used to adaptively increase the representativeness of minority class samples, alleviate data imbalance, and enable the multi-class machine learning model to learn features of all classes fairly. High-quality synthetic datasets are generated through deep learning via generative adversarial networks, which not only increases the number of samples but also improves data diversity, ensuring the comprehensiveness and complexity of the data during model training. A pre-defined decision classifier is used for data filtering, ensuring that the distribution of the generated synthetic data is similar to that of the actual predicted data. This improves the generalization ability of the final multi-class machine learning model, enabling it to perform better when facing new data. It significantly improves the classification accuracy of minority class customers, enhances the stability and accuracy of financial customer credit classification, and thus solves the technical problem of low classification accuracy for minority class customers in financial customer credit classification, which affects classification accuracy.

[0073] The following describes in detail another optional implementation method.

[0074] Figure 3 This is a schematic diagram of an optional financial customer credit classification method according to an embodiment of the present invention, such as... Figure 3 As shown, it includes:

[0075] Step 1: Collect historical datasets of bank customers and perform data preprocessing.

[0076] Obtain the raw dataset of bank customers, including various features such as basic information, credit history, and historical behavior.

[0077] Perform data cleaning on the dataset, including handling missing values ​​and removing outliers.

[0078] The numerical features of the data are standardized, and the categorical features are classified or isolated.

[0079] Step 2: Data balancing (using ADASYN oversampling to obtain a balanced dataset).

[0080] Input data: Original dataset ,in The feature vector corresponding to the customer's basic information. The labels are customer credit categories, where the categories are... For majority class, category , , It is a minority category.

[0081] For each minority class sample in any minority class Calculate its k nearest neighbor sample set, which includes majority class and minority class sample neighbors, where k can be set to 5.

[0082] ={ };

[0083] Calculation difficulty coefficient For each minority class sample Calculate the proportion of the majority class in its neighborhood to the total sample. :

[0084]

[0085] Number of synthesized samples: for each minority class sample Let the current total number of minority class samples be... The number of samples in the majority class is The total number of synthetic samples to be generated is Based on its difficulty level Proportionately allocate the number of samples generated per sample .

[0086] ;

[0087] .

[0088] Generate synthetic samples: randomly selected and its minority class nearest neighbor samples Generate random numbers ,exist and its minority class nearest neighbor samples Linear interpolation is performed in the middle to generate synthetic samples. .

[0089] ;

[0090] The final result is a balanced dataset with balanced class distributions. .

[0091] Step 3: Deep data synthesis (using the CTGAN network to obtain a high-quality verification dataset).

[0092] Balanced datasets that balance class distributions In the CTGAN deep learning network, the input data is a synthetic dataset whose generated data is equal to the number of balanced datasets. .

[0093] CTGAN is a generative adversarial network (GAN) model used to generate high-quality tabular data. It is suitable for processing datasets that combine customer credit classification features and numerical features. CTGAN includes normalization for numerical features to overcome non-Gaussian and multimodal distributions. It also incorporates a conditional generator trained by sampling to handle imbalanced discrete class features.

[0094] Step 4: Data distribution filtering.

[0095] Balance the dataset and synthetic datasets By merging, a dataset with a balanced data distribution is obtained. .

[0096] Will The original labels of the data are removed and set to 0. The labels of the customer dataset to be predicted are set to 1. The CatBoost classifier is then used for classification. CatBoost is a machine learning algorithm based on gradient boosting decision trees, which performs well in handling both categorical features and numerical values, and boasts high training speed and prediction accuracy. The generated synthetic data can be processed using the `predict_proba` method. The output is the probability of the predicted label. For each training sample, CatBoost outputs the probability P(y=1|) that the sample belongs to the data label to be predicted (y=1). ).

[0097] Finally, samples with a predicted probability of over 80% were selected. These samples are considered to be similar in distribution to the bank customer dataset that will be used for prediction.

[0098] Step 5: Predicting the credit category / rating of bank customers.

[0099] Filtered data As training data, a multilayer perceptron (MLP) model is used to classify the customers to be predicted into multiple categories. The basic parameters of the multilayer perceptron (MLP) are set as follows:

[0100] Network structure: It contains two hidden layers: input layer → 64-neuron hidden layer → 32-neuron hidden layer → output layer. The activation function of the hidden layer is ReLU, and the output layer uses the Softmax function.

[0101] Loss function: Weighted cross-entropy;

[0102] Regularization: L2 norm constraint (λ=0.01);

[0103] The final customer credit category output is obtained.

[0104] Through the embodiments of the present invention, the adaptive oversampling model ADASYN and the pre-set adversarial generative network CTGAN are used for data balancing and enhancement, and the CatBoost classifier is used for data filtering. Finally, by completing multi-class credit prediction, it has significant beneficial effects in dealing with data imbalance, improving the generalization ability of the model, and improving classification accuracy.

[0105] In bank customer credit classification tasks, customer data for certain credit categories is scarce, while samples from other categories constitute the majority, making it difficult for classification models to correctly learn the features of minority class samples. This invention uses the ADASYN adaptive synthetic oversampling technique to generate synthetic data for minority class samples, thereby balancing the class distribution of the dataset. The CTGAN network learns the distribution of real data through the mechanism of deep generative adversarial networks (GANs) to generate synthetic data that conforms to the features of the original data. Compared to traditional data augmentation methods, CTGAN can better capture the feature distribution of complex data, especially performing exceptionally well when dealing with data that mixes categorical and numerical features. By using CTGAN to generate high-quality synthetic samples, the diversity and coverage of the dataset are enhanced.

[0106] Furthermore, this invention constructs a binary classification problem by combining the generated data with real data labels. The CatBoost classifier is then used to filter the generated data, selecting those with a distribution similar to the data to be predicted. This effectively avoids the negative impact of low-quality or off-target samples on model training. The filtered dataset is more representative, which is helpful for the subsequent training and prediction of multi-class models, thus improving the final credit rating prediction accuracy.

[0107] The following is a detailed description with reference to another embodiment.

[0108] Example 2

[0109] The financial customer credit classification device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the foregoing method embodiment, and will not be repeated here.

[0110] Figure 4 This is a schematic diagram of an optional financial customer credit classification device according to an embodiment of the present invention, such as... Figure 4 As shown, the financial customer credit classification device may include: a dataset receiving unit 41, a data classification and filtering unit 42, and a credit classification unit 43.

[0111] Among them, the dataset receiving unit 41 is used to receive the basic information of financial customers and the credit dataset to be predicted.

[0112] The data classification and filtering unit 42 is used to classify and filter the target dataset and the credit dataset to be predicted using a preset decision classifier to form the final dataset. The final dataset is used to adjust the parameters of the multi-class machine learning model. The target dataset is a dataset with balanced data distribution obtained by merging the initial balanced dataset and the synthetic dataset. The initial balanced dataset is a dataset generated by adaptively synthesizing and oversampling minority class customer samples in the original dataset. The synthetic dataset is a dataset generated by using a preset generative adversarial network model to perform deep learning on the initial balanced dataset. The initial balanced dataset and the synthetic dataset have the same number of data.

[0113] Credit classification unit 43 is used to perform classification prediction of financial customer credit using an adjusted multi-class machine learning model and output the credit category of the financial customer.

[0114] The aforementioned financial customer credit classification device can receive basic information of financial customers and a credit dataset to be predicted through a dataset receiving unit 41. A data classification and filtering unit 42 uses a preset decision classifier to classify and filter the target dataset and the credit dataset to be predicted, forming a final dataset. The final dataset is then used to adjust the parameters of a multi-class machine learning model. The target dataset is a data distribution-balanced dataset obtained by merging an initial balanced dataset and a synthetic dataset. The initial balanced dataset is generated by adaptively synthesizing and oversampling minority class customer samples from the original dataset. The synthetic dataset is generated by performing deep learning on the initial balanced dataset using a preset generative adversarial network model. The initial balanced dataset and the synthetic dataset have the same number of data points. The credit classification unit 43 uses the adjusted multi-class machine learning model to perform financial customer credit classification prediction and output the credit category of the financial customer. In this embodiment, an adaptive synthetic oversampling strategy can be used to adaptively increase the representativeness of minority class samples, alleviate data imbalance, and enable the multi-class machine learning model to learn features of all classes fairly. High-quality synthetic datasets are generated through deep learning via generative adversarial networks, which not only increases the number of samples but also improves data diversity, ensuring the comprehensiveness and complexity of the data during model training. A pre-defined decision classifier is used for data filtering, ensuring that the distribution of the generated synthetic data is similar to that of the actual predicted data. This improves the generalization ability of the final multi-class machine learning model, enabling it to perform better when facing new data. It significantly improves the classification accuracy of minority class customers, enhances the stability and accuracy of financial customer credit classification, and thus solves the technical problem of low classification accuracy for minority class customers in financial customer credit classification, which affects classification accuracy.

[0115] Optionally, the financial customer credit classification device further includes, when generating the initial balanced dataset: a data preprocessing unit, used to preprocess the acquired original dataset of financial institution customers, wherein each piece of data in the original dataset includes at least: basic customer information, customer feature vector, and corresponding data label, and the data label includes at least: customer credit category; a nearest neighbor sample calculation unit, used to input the preprocessed original dataset into an adaptive synthetic oversampling model, which calculates K nearest neighbor samples for the minority class samples in the original dataset, wherein the minority class samples refer to samples in the original dataset whose data volume corresponding to the customer credit category is less than a preset data volume threshold, and K is a positive integer greater than 1; a generation quantity calculation unit, used to calculate the number of to-be-generated samples for each minority class sample in the original dataset based on the K nearest neighbor samples, and to interpolate between the minority class samples and the minority class nearest neighbor samples to generate the number of synthetic samples to be generated; and a balanced dataset generation unit, used to synthesize the synthetic samples corresponding to all minority class samples to generate the initial balanced dataset.

[0116] Optionally, the data preprocessing unit includes: a data cleaning module for cleaning the acquired raw dataset of financial institution customers, wherein the data cleaning method includes at least one of the following: missing value imputation and outlier removal; and a standardization module for standardizing the numerical features in the cleaned raw dataset.

[0117] Optionally, the generation quantity calculation unit includes: a difficulty coefficient calculation module, used to calculate the proportion of majority class samples in the K nearest neighbor samples of each minority class sample to the total samples, and determine the difficulty coefficient corresponding to the minority class sample based on the sample proportion; and a generation quantity allocation module, used to calculate the total number of synthetic samples to be generated for each minority class sample based on the total number of minority class samples and the number of majority class samples, and allocate the number of minority class samples to be generated based on the difficulty coefficient and the total number of synthetic samples.

[0118] Optionally, when generating a synthetic sample corresponding to the quantity to be generated, the quantity calculation unit performs linear interpolation between the minority class sample and the minority class nearest neighbor sample to generate a synthetic sample, it includes: a random selection module for randomly selecting a minority class sample and its minority class nearest neighbor sample; a random number generation module for generating a random number corresponding to the minority class sample and its minority class nearest neighbor sample; and a synthetic sample generation module for performing linear interpolation between the minority class sample and its minority class nearest neighbor sample based on the minority class sample, the minority class nearest neighbor sample, and the random number to generate a synthetic sample of the quantity to be generated.

[0119] Optionally, the data classification and filtering unit includes: a first label setting module, used to remove the data labels from each data sample in the target dataset and set the data labels to a first value; a second label setting module, used to set the data labels of each customer data sample in the credit dataset to be predicted to a second value, wherein the second value is different from the first value; a label prediction probability output module, used to output the label prediction probability corresponding to each data sample in the target dataset by a preset decision classifier using the target dataset with the labels set and the credit dataset to be predicted; and a dataset classification and filtering module, used to filter out data samples with label prediction probabilities greater than a preset probability threshold to obtain the final dataset.

[0120] Optionally, when the multi-class machine learning model is a multilayer perceptron model, it includes: a neural network architecture with two hidden layers, the first hidden layer including a first set of neurons, the second hidden layer including a second set of neurons, all hidden layers using the ReLU activation function, wherein the number of neurons in the first set is greater than the number of neurons in the second set; an output layer using the Softmax activation function; wherein, weighted cross-entropy is used as the loss function during the training of the multi-class machine learning model.

[0121] The aforementioned financial customer credit classification device may also include a processor and a memory. The aforementioned dataset receiving unit 41, data classification and filtering unit 42, credit classification unit 43, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0122] The processor described above contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters allows for accurate classification of customer credit categories.

[0123] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0124] Example 3

[0125] Embodiments of this application may provide an electronic device. Figure 5 This is a structural block diagram of an electronic device that performs a method for classifying the credit of financial customers according to an embodiment of this application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (Only one is shown) processor 502, memory 504, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0126] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the financial customer credit classification method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned financial customer credit classification method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0127] The processor can access information and applications stored in memory via a transmission device to perform the following steps: receiving basic information about financial customers and a dataset of credit data to be predicted; classifying and filtering the target dataset and the dataset of credit data to be predicted using a preset decision classifier to form a final dataset; adjusting the parameters of a multi-class machine learning model using the final dataset. The target dataset is a dataset with a balanced data distribution obtained by merging an initial balanced dataset and a synthetic dataset. The initial balanced dataset is generated by adaptively oversampling minority class customer samples from the original dataset. The synthetic dataset is generated by performing deep learning on the initial balanced dataset using a preset generative adversarial network model. The initial balanced dataset and the synthetic dataset have the same number of data points. The adjusted multi-class machine learning model is then used to perform classification prediction of the financial customer's credit and output the financial customer's credit category.

[0128] Those skilled in the art will understand that Figure 5 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 5 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 5 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 5 The different configurations shown.

[0129] Those skilled in the art will understand that all or part of the steps in the various financial customer credit classification methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0130] Example 4

[0131] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the financial customer credit classification method provided in Embodiment 1.

[0132] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the financial customer credit classification method of any one of the above embodiments.

[0133] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0134] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the financial customer credit classification method described in various embodiments of this application.

[0135] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the financial customer credit classification method described in various embodiments of this application.

[0136] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0137] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0138] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0140] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0141] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0142] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of classifying credit of a financial customer, characterized by, The method comprises the following steps: receiving basic information of a financial customer and a credit data set to be predicted; using a preset decision classifier to classify and filter a target data set and the credit data set to be predicted, forming a final data set, and using the final data set to adjust parameters of a multi-class machine learning model, wherein the target data set is a data distribution balanced data set obtained by merging an initial balanced data set and a synthetic data set, the initial balanced data set is a data set generated by adaptively synthesizing oversampling of a minority class customer sample in an original data set, and the synthetic data set is a data set generated by deep learning of the initial balanced data set using a preset generative adversarial network model, and the data quantity of the initial balanced data set and the synthetic data set is consistent; using the adjusted multi-class machine learning model to perform classification prediction of the credit of the financial customer, and outputting the credit category of the financial customer.

2. The classification method of claim 1, wherein, When the initial balanced data set is generated, the method comprises the following steps: preprocessing an original data set of a financial institution customer, wherein each piece of data in the original data set at least includes customer basic information, a customer feature vector, and a corresponding data label, and the data label at least includes a customer credit category; inputting the preprocessed original data set into an adaptive synthetic oversampling model, calculating K neighbor samples of a minority class sample in the original data set by the adaptive synthetic oversampling model, wherein the minority class sample refers to a sample in the original data set whose corresponding data quantity of the customer credit category is less than a preset data quantity threshold, and K is a positive integer greater than 1; based on the K neighbor samples, calculating the number of synthetic samples to be generated for each minority class sample in the original data set, and interpolating between the minority class sample and the minority class nearest neighbor sample to generate the number of synthetic samples to be generated; generating the initial balanced data set by synthesizing all the synthetic samples corresponding to the minority class samples.

3. The classification method of claim 2, wherein, The preprocessing step of the original data set of the financial institution customer obtained comprises the following steps: data cleaning of the original data set of the financial institution customer obtained, wherein the data cleaning method comprises at least one of the following: missing value supplement and abnormal value removal; standardizing the numerical features in the cleaned original data set.

4. The classification method of claim 2, wherein, The step of calculating the number of synthetic samples to be generated for each minority class sample in the original data set based on the K neighbor samples comprises the following steps: calculating the proportion of majority class samples in the K neighbor samples of each minority class sample, and determining a difficulty coefficient corresponding to the minority class sample based on the sample proportion; for each minority class sample, calculating the total number of synthetic samples to be generated based on the total number of minority class samples and the number of majority class samples, and allocating the number of synthetic samples to be generated for each minority class sample based on the difficulty coefficient and the total number of synthetic samples.

5. The classification method of claim 2, wherein, The step of interpolating between the minority class sample and the minority class nearest neighbor sample to generate the synthetic samples corresponding to the number of synthetic samples to be generated comprises the following steps: randomly selecting a minority class sample and a minority class nearest neighbor sample of the minority class sample; generating a random number corresponding to the minority class sample and the nearest neighbor sample of the minority class; performing linear interpolation between the minority class sample and the nearest neighbor sample of the minority class based on the minority class sample, the nearest neighbor sample of the minority class, and the random number to generate the to-be-generated number of synthetic samples.

6. The classification method of claim 1, wherein, The step of classifying and screening the target data set and the to-be-predicted credit data set by using a preset decision classifier to form a final data set includes: removing the data label of each data sample in the target data set and setting the data label as a first value; setting the data label of each customer data sample in the to-be-predicted credit data set as a second value, wherein the second value is different from the first value; outputting, by the preset decision classifier, the label prediction probability corresponding to each data sample in the target data set by using the target data set and the to-be-predicted credit data set with the set labels; screening out data samples with a label prediction probability greater than a preset probability threshold to obtain the final data set.

7. The classification method of claim 1, wherein, When the multi-class machine learning model is a multi-layer perception model, the method includes: a neural network architecture with two hidden layers, a first hidden layer including a first set number of neurons and a second hidden layer including a second set number of neurons, and all hidden layers using a ReLU activation function, wherein the first set number is greater than the second set number; an output layer using a Softmax activation function; wherein a weighted cross-entropy is used as a loss function during training of the multi-class machine learning model.

8. An apparatus for classifying credit of a financial customer, characterized by comprising: The method includes: a data set receiving unit configured to receive basic information of a financial customer and a to-be-predicted credit data set; a data classification and screening unit configured to classify and screen the target data set and the to-be-predicted credit data set by using a preset decision classifier to form a final data set, and to adjust parameters of a multi-class machine learning model by using the final data set, wherein the target data set is a data distribution balanced data set obtained by merging an initial balanced data set and a synthetic data set, the initial balanced data set is a data set generated by performing adaptive synthetic oversampling on minority class customer samples in an original data set, and the synthetic data set is a data set generated by performing deep learning on the initial balanced data set by using a preset generative adversarial network model, and the data quantity of the initial balanced data set is consistent with that of the synthetic data set; a credit classification unit configured to perform classification prediction of a financial customer credit by using the adjusted multi-class machine learning model and output a credit category of the financial customer.

9. An electronic device, comprising: The computer program is executed by a processor to implement the steps of the method for classifying a financial customer credit according to any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the method for classifying a financial customer credit according to any one of claims 1 to 7.

Images