Client classification method, device and equipment based on industrial chain, storage medium and program product

Abstract

The embodiment of the invention provides a customer classification method and device based on an industrial chain, equipment, a storage medium and a program product, and relates to the field of financial science and technology or other related fields. The method comprises the following steps: acquiring credit data of a to-be-classified target customer; the credit data comprises basic information data, loan behavior data and industrial chain associated data of the target customer; performing feature extraction on the credit data of the target customer based on chi-square test to obtain a feature vector corresponding to the credit data; inputting the feature vector into a pre-constructed random forest model to obtain a classification result of the target customer; the random forest model is optimized based on an improved spotted serow algorithm, and the improved spotted serow algorithm optimizes the random forest model through at least one of collaborative optimization of a chaotic initialized population, a self-adaptive fitness function, the number of trees and the number of leaf nodes. According to the method, the customer classification accuracy based on the industrial chain is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of financial technology or other related fields, and in particular, relates to a customer classification method and device based on an industrial chain, equipment, a storage medium and a program product. BACKGROUND

[0002] With the growth of credit demand of small and micro enterprises and long-tail customers, financial institutions such as banks need to identify the industrial chain category of customers through data analysis to optimize the allocation of credit resources and reduce the cost of risk control.

[0003] There are mainly two implementation methods for traditional customer classification technology based on industrial chain: one is a classification method based on expert rules, which manually formulates industrial chain classification standards (such as enterprise operating range keyword matching); the other is a classification method based on traditional machine learning, such as using a random forest model for classification, but the model hyperparameters (such as the number of trees and the number of leaf nodes) are usually optimized by grid search or random search, which has the problems of unreasonable parameter combination and large calculation overhead. Therefore, some solutions introduce intelligent optimization algorithms.

[0004] However, when applying the above classification technology to classify inclusive credit customers, there is still the defect of low accuracy. SUMMARY

[0005] The present application provides a customer classification method and device based on an industrial chain, equipment, a storage medium and a program product to solve the defect of low accuracy in customer classification based on an industrial chain in the known technology.

[0006] In a first aspect, the present application provides a customer classification method based on an industrial chain, which comprises:

[0007] Obtaining credit data of a target customer to be classified; the credit data includes basic information data, loan behavior data and industrial chain association data of the target customer;

[0008] Performing feature extraction on the credit data of the target customer based on a chi-square test to obtain a feature vector corresponding to the credit data;

[0009] Inputting the feature vector into a pre-constructed random forest model to obtain a classification result of the target customer; the random forest model is optimized based on an improved spotted hyena algorithm, and the improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of population, self-adaptive fitness function, number of trees and number of leaf nodes in cooperative optimization.

[0010] In a second aspect, the present application provides a customer classification device based on an industrial chain, which comprises:

[0011] An acquisition module is configured to acquire credit data of a target customer to be classified, wherein the credit data comprises basic information data, loan behavior data and industry chain association data of the target customer;

[0012] An extraction module is configured to perform feature extraction on the credit data of the target customer based on a chi-square test, to obtain a feature vector corresponding to the credit data of the target customer;

[0013] An analysis module is configured to input the feature vector into a pre-constructed random forest model, to obtain a classification result of the target customer, wherein the random forest model is optimized based on an improved spotted hyena algorithm, and the improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of a population, self-adaptive fitness function, number of trees and number of leaf nodes in collaborative optimization.

[0014] In a third aspect, the present application provides an electronic device, comprising at least one processor, and a memory connected to the processor in communication;

[0015] The memory stores computer-executable instructions;

[0016] The at least one processor executes the computer-executable instructions stored in the memory, to implement the method according to any one of the first aspect.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method according to any one of the first aspect.

[0018] In a fifth aspect, the present application provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor to implement the method according to any one of the first aspect.

[0019] The application provides a customer classification method based on an industrial chain. For any target customer to be classified, feature extraction is performed on obtained credit data of the target customer based on chi-square test to obtain a feature vector corresponding to the credit data. Then, the feature vector is input into a pre-constructed random forest model to obtain a classification result of the target customer. In the process, on the one hand, the credit data contains industrial chain correlation data, and the random forest model used is optimized based on an improved spotted hyena algorithm, which expands the feature search space by chaotic initialization to cover high-dimensional attributes of the industrial chain, or strengthens the screening of key features of the industrial chain by an adaptive fitness function, or adapts to the hierarchical features of the industrial chain by collaborative optimization of tree parameters, so that the model can accurately capture the industrial chain attributes of the customer and improve the classification accuracy. On the other hand, chi-square test is used for feature extraction, which meets the demand for screening the effectiveness of data features in the inclusive finance scenario, can eliminate redundant features to reduce the computational complexity, and can maintain population diversity in the early evolution stage based on the improved spotted hyena algorithm to avoid premature convergence of the algorithm, thereby accelerating the parameter optimization efficiency of the model, thereby being beneficial to efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are incorporated herein and form a part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0021] Figure 1 An application scenario diagram of a customer classification method based on an industrial chain provided by an embodiment of the application;

[0022] Figure 2 A flowchart of a customer classification method based on an industrial chain provided by an embodiment of the application Figure 1 ;

[0023] Figure 3 A flowchart of a customer classification method based on an industrial chain provided by an embodiment of the application Figure 2 ;

[0024] Figure 4 A structural diagram of a customer classification device based on an industrial chain provided by an embodiment of the application;

[0025] Figure 5 A structural diagram of an electronic device provided by an embodiment of the application.

[0026] Through the above drawings, the specific embodiments of the application have been shown, and will be described in more detail hereinafter. These drawings and the written description are not intended to limit the scope of the inventive concept in any way, but to illustrate the inventive concept to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0027] Exemplary embodiments will be described in detail herein with reference to the drawings. When the description is made with reference to the drawings, the same or similar components are designated with the same or similar reference numerals throughout the several drawings. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present application. Instead, they only represent examples of devices and methods consistent with some aspects of the present application, as detailed in the appended claims.

[0028] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards of relevant countries and regions, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal.

[0029] It should be noted that the customer classification method, device, equipment, storage medium and program product based on the industrial chain provided by the present application can be used in the field of financial technology, and can also be used in any field other than financial technology. The application field of the customer classification method, device, equipment, storage medium and program product based on the industrial chain in the present application is not limited.

[0030] In traditional bank credit business, accurate customer classification and marketing are the core links of risk control and business expansion. With the growth of credit demand of small and micro enterprises and long-tail customers, banks and other financial institutions not only need to judge whether customers are high-quality customer groups, but also need to identify their industrial chain categories through data analysis to optimize credit resource allocation and reduce risk control costs.

[0031] However, the existing technology has double defects in the analysis of the industrial chain dimension: on the one hand, traditional classification technology is difficult to meet the demand of in-depth analysis of the industrial chain. The method based on expert rules sets the classification standard manually (such as matching the key words of enterprise business scope), which is subjective and cannot dynamically adapt to the changes of industrial policy; the method based on traditional machine learning (such as random forest) can process massive data, but the model hyperparameters (such as the number of trees and the number of leaf nodes) are often optimized by grid search or random search, which has the problems of unreasonable parameter combination and large calculation overhead. Although some solutions introduce intelligent optimization algorithms, the accuracy of classifying small and micro enterprises in the industrial chain in the inclusive credit scene is still significantly insufficient, and it is difficult to capture the hierarchical relationship implied in the industrial chain (such as the risk difference between core enterprises and upstream and downstream supporting enterprises).

[0032] On the other hand, existing algorithms are inefficient in high-dimensional credit data processing. For example, the K-Nearest Neighbor Algorithm (KNN) needs to store all training data and calculate all neighbor samples. In the processing of high-dimensional credit data containing basic information, loan behavior and industry chain correlation data, the requirement for computing resources is extremely high, and the classification accuracy in high-dimensional feature space is significantly reduced, which cannot effectively identify the industry chain attributes (such as industry type and supply chain position) of customers, resulting in great uncertainty in credit investment and difficulty in adapting to dynamic industry policies and market demand.

[0033] In summary, due to the lack of deep feature extraction of industry chain data, efficient hyperparameter optimization mechanism and high-dimensional data processing capability, the existing technology faces the dual challenges of "low accuracy and poor computing efficiency" in the industry chain classification of inclusive credit customers, and there is an urgent need for a customer classification method that integrates industry chain feature analysis, intelligent algorithm optimization and efficient data processing.

[0034] Therefore, the present application provides a customer classification method based on an industry chain, which aims to solve the above technical problems of known technologies. Specifically, the present application proposes to optimize a random forest model based on an improved spotted hyena algorithm, and to obtain a classification result by using the random forest model and credit data containing industry chain correlation data. Specifically, the improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of population, adaptive fitness function, number of trees and number of leaf nodes in cooperative optimization.

[0035] It should be understood that the method of the present application can be applied not only in the inclusive credit scenario, but also in the risk assessment of upstream and downstream enterprises in supply chain finance, customer credit stratification in consumer finance, farmer loan review in agricultural industry chain, cross-border financing risk control of cross-border trade enterprises, and credit support for small and medium-sized enterprises in emerging industries. These scenarios can achieve accurate classification of customers in different fields by integrating customer basic information, credit behavior data and industry chain correlation features, using the chi-square test and the random forest model optimized by the improved spotted hyena algorithm, which not only meets the demand for deep analysis of industry chain data in various scenarios, but also improves the efficiency of high-dimensional data processing and classification accuracy through intelligent algorithms. For example, Figure 1 An application scenario diagram of a customer classification method based on an industry chain provided by an embodiment of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the method of the present application can be used in the inclusive credit scenario, and the method of the present application is executed by an independently deployed electronic device.

[0036] Specifically, in this scenario, the inclusive credit business terminal interacts with the electronic device to obtain the credit data of the target customer to be classified from the inclusive credit business terminal, and performs feature extraction on the credit data based on the chi-square test to obtain a feature vector, and then inputs the feature vector into a pre-constructed random forest model to obtain the classification result of the target customer and feeds back to the inclusive credit business terminal.

[0037] In the above process, since the random forest model adopted by the electronic device is optimized based on the improved spotted hyena algorithm, the algorithm expands the feature search space through chaotic initialization of the population, or combines an adaptive fitness function to strengthen the screening of key features of the industrial chain (such as core enterprise cooperation frequency and supply chain level), or optimizes the number of trees and the number of leaf nodes of the random forest in coordination, so that it adapts to the high dimensionality and hierarchical features of the industrial chain data of inclusive credit small and micro enterprises, so that the model can more accurately capture the attribute features of the customer in the industrial chain and improve the classification accuracy; at the same time, the chi-square test eliminates redundant features to reduce the computational complexity, and the improved algorithm avoids local convergence through dynamic adjustment to speed up the parameter optimization efficiency, and finally realizes the dual improvement of classification accuracy and processing efficiency in the inclusive credit scenario.

[0038] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0039] The present embodiment provides a customer classification method based on an industrial chain. The method of the present application will be described in detail below by taking the application in the inclusive credit scenario as an example. Figure 2 The flowchart of a customer classification method based on an industrial chain provided by the present embodiment Figure 1 As shown in the figure, Figure 2 The method of the present embodiment includes:

[0040] S201, obtaining the credit data of the target customer to be classified.

[0041] The credit data includes the basic information data, loan behavior data and industrial chain association data of the target customer.

[0042] Specifically, in this embodiment, the basic information data includes the number of loan applications, the number of application platforms, the application amount, the credit amount, the loan amount, the repayment history, the overdue history, the number of overdue times, the collection effect, etc.; the loan behavior data is extracted based on the query, operation and other behavior point data of the customer, mainly including the number of times of opening, registering and logging in the platform, the number of times of clicking, applying, increasing the amount, authorizing and other loan-related operations, the click frequency, the page dwell time, etc.; the industry chain association data includes the industry to which the customer belongs, the asset size of the customer, the information of the associated enterprises of the customer, the industry proportion, the credit information of the upstream and downstream enterprises, the industry chain data, the industrial park data, the industrial incentive, etc.

[0043] In this embodiment, the electronic device obtains the encrypted credit data from the terminal corresponding distributed database based on the unique identification (such as customer ID) of the target customer by establishing a data interface with the inclusive credit business terminal, and performs desensitization processing. It should be understood that at least two target customers may need to be classified in the same period, and the electronic device in this embodiment processes each target customer in parallel. In actual application, considering the actual computing power of the electronic device, each target customer can also be processed in sequence, and when the number of customers to be processed exceeds the concurrent processing threshold (such as the number of CPU cores x 2) of the electronic device, the serial processing mode based on the task scheduling algorithm can be automatically switched to, so as to guarantee the single customer processing time (such as ≤3 seconds / customer), which is not limited in this embodiment.

[0044] In this embodiment, the industry chain association data of the target customer is obtained, which is used in subsequent classification analysis, can place the customer in the global perspective of the industry chain, combined with the credit information of the upstream and downstream enterprises, the industry proportion, the industrial incentive and other multi-dimensional characteristics, deeply excavates the role positioning and risk attribute of the customer in the industry chain, and then uses the random forest model optimized by the improved spotted hyena algorithm to accurately quantify the risk resistance ability and credit value of the customer, reduces the information asymmetry, assists the bank in formulating differentiated credit strategies, and significantly improves the accuracy of customer classification and the scientificity of credit decision.

[0045] S202, feature extraction is performed on the credit data of the target customer based on the chi-square test to obtain a feature vector corresponding to the credit data.

[0046] In this embodiment, after obtaining the credit data, the electronic device first pre-processes the credit data, and then extracts features from the pre-processed credit data to obtain a feature vector corresponding to the credit data. In this embodiment, the electronic device pre-processes the credit data as follows to obtain the processed credit data: outlier processing, missing value processing, and data merging processing.

[0047] Specifically, in this embodiment, the outlier processing is implemented through the following process: based on the Z-score method, the outliers are judged and processed, and the outliers are removed by The obtained data is standardized, the absolute value of the distance between the data points and the mean value is calculated, and if the difference is not less than 3 standard deviations, it is marked as a missing value, and the missing value is processed. Among them, is the standardized data, is the mean value of the data set, , is the standard deviation of the data set, .

[0048] Specifically, in this embodiment, the missing value processing is realized through the following process: according to the data after the abnormal value processing, the number and proportion of missing values are viewed, when the missing rate of the feature corresponding to the target customer data is lower than the specified threshold and the missing values appear randomly in the sample, the mean, median or mode is filled, when the missing rate of the feature is greater than the specified threshold and the explanatory of the variable is not high related to the cluster, the feature is directly deleted. It should be understood that the specified threshold of this part can be set according to the actual situation.

[0049] Specifically, in this embodiment, the data merging processing specifically refers to the merging of different source data. For basic information data, loan behavior data, and industry chain association data, the electronic device integrates the data scattered in different databases (such as credit system, behavior log library, and industry database) into structured data sets based on the unique identification (such as customer ID) of the target customer through data cleaning, format unification, and redundant data elimination, ensuring that the subsequent chi-square test can accurately evaluate the correlation between each feature and the customer classification label, avoiding feature selection bias caused by inconsistent data, and providing high-quality input data for the random forest model optimized by the improved spotted hyena algorithm.

[0050] In actual application, the electronic device can perform at least one of the above-mentioned pre-processing on the credit data, which is not limited in this embodiment.

[0051] After obtaining the pre-processed credit data, the electronic device extracts features from the pre-processed credit data to obtain a feature vector. Since there are a large number of redundant features in the credit data that are irrelevant to customer classification, the chi-square test can measure the independence of features and classification labels through statistical quantities, thereby selecting key features that have a significant impact on classification results. Therefore, in this embodiment, the electronic device extracts features from the pre-processed credit data based on the chi-square test.

[0052] It should be understood that the chi-square test measures the independence of feature values and classification labels (such as "high-quality customers" / "non-high-quality customers") to judge the contribution of the feature to classification. The core idea is: if the feature and the classification label have significant correlation, the feature has important value for the classification result, otherwise it is considered as a redundant feature.

[0053] On this basis, in the embodiment, the electronic device obtains the feature vector in the following manner: a contingency table is constructed based on the category label and each data feature in the credit data, and actual observation values and theoretical values of each cell of the contingency table are calculated; the actual observation values and the theoretical values are substituted into a chi-square test formula to calculate chi-square values of the data features; the data features are arranged in descending order of the chi-square values to obtain a feature importance list, and a target feature is selected according to the feature importance list to generate the feature vector; the chi-square value of the target feature satisfies at least one of the following conditions: greater than a preset chi-square value, and ranking greater than a preset ranking.

[0054] More specifically, the target feature obtained through the screening is standardized or normalized to obtain a processed target feature; and the processed target features are arranged in a selected feature order to form a vector to obtain the feature vector.

[0055] In the embodiment, the electronic device constructs a 2*2 contingency table (four-cell table) for a binary classification label, counts actual observation values (for example, A, B, C, and D represent the cross-counts of the presence / absence of a feature and the positive / negative examples of a label) of each data feature and label combination, and calculates theoretical values based on the “feature and label independence” hypothesis; then the chi-square formula is used to calculate the chi-square value of the feature, and the greater the value, the stronger the association between the feature and the binary classification label, and then the target feature is selected in descending order of the chi-square value; where N is the total number of data features, and AD-BC is the difference between the actual observation value and the theoretical value.

[0056] The electronic device extends the binary classification chi-square test to multi-classification contingency table analysis for multi-classification labels (for example, “superior customer”, “high-quality customer”, “general customer”): first, an m*k contingency table is constructed for each data feature (m is the number of feature values, and k is the number of label categories), and the actual observation values and theoretical values of each cell are calculated; then the multi-classification chi-square formula is used to calculate the chi-square value of the feature, which reflects the overall discrimination ability of the feature for k categories; finally, the target feature is selected in descending order of the chi-square value, for example, if the chi-square value of “monthly income interval” is significantly higher than “whether there is a credit card”, the former is more suitable for distinguishing customers of different categories; where O ij is the actual observation value, E ij is the theoretical value, which is calculated by the formula Row i is the total of the i-th row, col j is the total of the j-th column, and N is the total number of data features.

[0057] Further, after obtaining the target features, the electronic device converts features of different dimensions (such as "income amount" in units of "ten thousand yuan" and "debt ratio" in units of "percentage") into the same scale (such as the interval [0, 1] or a distribution with a mean of 0 and a standard deviation of 1) based on a feature standardization or normalization method, to avoid model training bias caused by differences in feature dimensions; at the same time, the processed features are arranged in order according to business logic or algorithm output, to form a feature vector with consistent dimensions.

[0058] Through the above settings, whether it is a binary classification or a multi-classification scenario, the electronic device realizes statistical significance screening of features through chi-square test, and improves the model adaptability of the feature vector through standardization processing. Specifically, on the one hand, it can improve the training efficiency of the subsequent credit customer classification model, and avoid computational redundancy caused by too high feature dimensions; on the other hand, it can enhance the interpretability of the model - features with high chi-square values can be directly used as key criteria for judging customer categories, such as whether there is a stable income, and the contribution of such features to distinguishing "high-quality customers" can be intuitively reflected through chi-square values, providing data support for credit risk control decisions.

[0059] S203, inputting the feature vector into a pre-constructed random forest model to obtain a classification result of the target customer.

[0060] The random forest model is optimized based on an improved spotted hyena algorithm. The improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of the population, self-adaptive fitness function, and cooperative optimization of the number of trees and the number of leaf nodes.

[0061] In this embodiment, the pre-constructed random forest model is trained through the following process: based on the improved spotted hyena algorithm, the original sample data is subjected to feature screening and parameter optimization to generate a preprocessed data set suitable for the random forest model; the preprocessed data set is divided into k subsets with balanced class proportions by using stratified k-fold cross-validation, and the random forest model is trained using k-1 subsets and validated using 1 subset, and the model training process is optimized through comprehensive evaluation results after multiple iterations to obtain the final random forest model.

[0062] Specifically, the pre-processed data set includes multiple groups of input feature vectors and corresponding classification labels. In this embodiment, the classification labels include five levels of special customers, high-quality customers, general customers, cautious customers, and rejected customers. The core determination conditions of each level are as follows: the special customer indicated customer needs to meet the conditions of “legal representative age 30-60 years old, belonging to the national key support industry or strategic emerging industry or overall high-scale loan industry, enterprise age 5 years or more, owning leading technology patents, fixed operating site located in the national key development area, and good credit record”; the high-quality customer indicated customer needs to meet the conditions of “advanced manufacturing industry or green low-carbon industry or medium-scale loan industry, fixed operating site, good credit record, loan record, and stable upstream and downstream enterprises”; the general customer indicated customer needs to meet the conditions of “enterprise age 2 years or more, information technology industry or financial real estate industry, good credit record, less than 3 customer application platforms, and fixed operating site”; the cautious customer indicated customer needs to meet the conditions of “enterprise age less than 2 years, credit record with overdue record, legal representative age less than 30 or more than 60, loan application in more than 3 institutions, and sunset industry”; and the rejected customer indicated customer needs to meet the conditions of “enterprise age less than half a year, no fixed operating site, bad credit record, multiple risks, and blacklisted customer”.

[0063] It should be understood that the electronic device trains the random forest model based on the improved spot hyena algorithm using the pre-processed data set, and takes the corresponding classification label as the training target when the input feature vector is input.

[0064] More specifically, the electronic device uniformly divides the pre-processed data set into k non-overlapping subsets, and maintains the same class proportion in each subset as in the pre-processed data set. For example, if the ‘special customer’ accounts for 20% in the original data, the ‘special customer’ proportion in each subset is also about 20%. During the iterative training process, one subset is selected as the validation set, and the remaining k-1 subsets are used as the training set to train the random forest model based on the improved spot hyena algorithm. After multiple rounds of iteration, the evaluation results on each round of validation set are averaged to obtain the comprehensive evaluation of the model performance.

[0065] The customer classification method based on the industrial chain provided in this embodiment can significantly improve the processing efficiency while improving the classification accuracy by fusing the industrial chain related data, using the chi-square test to select key features, and using the random forest model optimized by the improved spot hyena algorithm. On the one hand, the multi-dimensional feature system and algorithm optimization make the model more accurate in identifying customer categories, especially in complex industrial chain scenarios, which can deeply mine customer risk attributes and credit value. On the other hand, the feature engineering filters a large amount of redundant information, and cooperates with the parallel processing and dynamic scheduling mechanism to greatly improve the data processing speed, realizing efficient and accurate customer classification and providing accurate and efficient decision support for inclusive credit business.

[0066] As a further illustration, the application also provides an embodiment of an industry chain-based customer classification method for describing the detailed process of optimizing the random forest algorithm based on the improved spotted hyena algorithm. Figure 3 The flowchart of an industry chain-based customer classification method provided by the embodiment of the application Figure 2 As shown in Figure 3 The method of the embodiment includes:

[0067] S301, initializing the population of the random forest model using a chaotic mapping to generate a traversal sequence, to obtain an initialized population.

[0068] Each individual included in the initialized population is used to represent a set of parameter combinations, including the number of trees and the number of leaf nodes.

[0069] In the embodiment, the electronic device uses a Tent mapping function to generate the initialized population. The specific process is as follows: first, determine the search space of the number of trees and the number of leaf nodes; generate a chaotic sequence in a two-dimensional Euclidean space through the Tent mapping function; the expression of the Tent mapping function is: wherein a is a control parameter, the value range is 0-1, d is used to represent the dimension, the value is 2, y i，d is used to represent the chaotic value of the ith individual in the dth dimension, and the value range is also 0-1; map the chaotic sequence to the search space through linear transformation to generate the initialized population.

[0070] In the embodiment, the search space of the number of trees is configured as [50, 200], and the search space of the number of leaf nodes is configured as [1, 20]. It should be understood that in actual application, it can also be configured as other ranges, which are not limited in the embodiment.

[0071] In the embodiment, a is specifically 0.5, at which time the mapping formula is represented as In the embodiment, the Tent mapping function is used to obtain the initialized population, which can utilize the traversal and randomness of the chaotic sequence to make the generated parameter combinations (such as the number of trees and the number of leaf nodes) uniformly distributed in the search space, avoiding the problem of population aggregation in the local area caused by traditional random initialization. For example, when the search space of the number of trees is [50, 200], the chaotic value generated by the Tent mapping can ensure that the number of trees in the initialized population is approximately uniformly distributed between 50 and 200, thereby covering more potential optimal parameter combinations, providing more initial search samples for the global optimization of the improved spotted hyena algorithm, and effectively improving the efficiency and accuracy of subsequent random forest model parameter optimization.

[0072] In practical applications, the electronic device can also use other mapping functions to obtain the initialization population, such as a Logistic mapping function or a Sine mapping function, adjust the mapping parameters to make the chaotic sequence uniformly distributed in the interval [0, 1], and then map the chaotic value to the preset parameter search space through linear transformation, to ensure that the initialization population has good ergodicity in the entire search range, avoid the population aggregation problem caused by traditional random initialization, and lay a global search foundation for subsequent algorithm optimization. In the present embodiment, this is not limited.

[0073] S302, the fitness function is used to calculate the individual fitness of each parameter combination in the initialization population.

[0074] The fitness function is constructed based on at least one of the model classification accuracy, the calculation overhead, and the population parameter distance, so that individuals far from the population mean have a higher selection probability.

[0075] S303, iteratively update each parameter combination until the maximum iteration number is reached or the individual fitness value converges, output the optimal parameter combination, and obtain the pre-constructed random forest model.

[0076] It should be understood that after obtaining the initialization population, the population parameters need to be initialized, including the number of population individuals, the maximum iteration number, the initial individual fitness, and the individual position. The initial individual fitness is calculated by an internal 5-fold cross-validation strategy.

[0077] Specifically, assuming that the number of population individuals of the initialization population is N, the initialization population parameters are randomly generated N individual positions, and the position of the i-th individual is wherein, Tni represents the number of trees at the current position of individual i, Ln i represents the number of leaf nodes at the current position of individual i.

[0078] On this basis, the electronic device first calculates the individual fitness of each individual in the current population based on the dynamically adaptive fitness function; the dynamically adaptive fitness function is represented as wherein, is the individual fitness of individual i in the t-1 iteration process, is the individual fitness corresponding to the optimal individual in the t-1 iteration process, is the individual fitness of individual i in the t iteration process, t is the current iteration number, and T is the maximum iteration number.

[0079] updating the individual positions of at least part of the individuals in the current population according to an update formula to obtain an updated current population; the update formula is expressed as: wherein, is the updated position of individual i, is the position of individual i in the search space corresponding to the jth index in the last iteration process, j is used to represent the number of trees or the number of leaf nodes, , is the search upper bound of , , , is a random number in the range of 0-1.

[0080] When the maximum number of iterations is reached, and / or, the individual fitness of the optimal individual changes less than a preset threshold for continuous k iterations, the updated current population is taken as the optimal parameter combination, and a random forest model is obtained according to the optimal parameter combination.

[0081] It should be understood that in the first iteration process, the current population is the initialized population.

[0082] In this embodiment, when the individual fitness value is less than 0.6T, linear scaling is performed; when the individual fitness value is greater than or equal to 0.6T, a tangent function is used for nonlinear scaling to increase the selection pressure, so that the algorithm is more inclined to select better individuals in the later period.

[0083] Through the method of this embodiment, the global coverage of the chaotic initialized population on the parameter space, the dynamic adjustment of the adaptive fitness function on the search direction, and the precise control of the model complexity by the synergistic optimization of the number of trees and the leaf nodes, all three work together to make the random forest model realize multi-dimensional optimization in the inclusive credit scene: the ergodicity of the chaotic sequence ensures that the initial parameters cover more potential optimal solutions, the adaptive function avoids the algorithm falling into local optimum, and the parameter synergistic optimization balances the model generalization ability, finally improves the accuracy and robustness of customer classification, and the synergistic effect of the three optimization methods can make the model more accurately identify the customer risk attributes and credit value in complex industrial chain data, and provide reliable support for differentiated credit strategy.

[0084] As another design, in practical applications, the improved spotted hyena algorithm can optimize the random forest model by at least one of the following, and any optimization can improve the classification accuracy:

[0085] Only chaotic initialization population: without adaptive function or synergistic optimization, only through Logistic mapping to generate uniformly distributed initial parameters, which can make the classification accuracy of the model on credit data improve by 5%-10% compared with traditional random initialization, because the initial parameters cover more potential optimal solutions;

[0086] Adaptive fitness function only: Without using chaotic initialization, only dynamically adjusting the weights of accuracy and parameter distance in the fitness function (e.g., the distance weight decreases from 0.2 to 0.1 as the number of iterations increases), the algorithm can avoid getting trapped in local optima, thus improving the recognition rate of minority classes such as "cautious customers" by more than 15%.

[0087] Co-optimization of tree number and leaf nodes only: without involving chaos or adaptive functions, only synchronously searching for the optimal combination of Tn and Ln, which reduces the model generalization error by 8%-12% compared to optimizing Tn alone (fixing Ln=5), because the synergistic effect between parameters is fully exploited.

[0088] This application does not limit the optimization method, and supports flexible selection of at least one optimization method such as chaotic initialization, adaptive fitness function, or parameter co-optimization according to actual needs. All of these methods can improve the model classification accuracy and provide an efficient and accurate solution for credit customer classification. Developers can flexibly choose based on computing resources (e.g., prioritizing co-optimization when computing power is limited) or business needs (e.g., using a combination of chaos and adaptive optimization when pursuing the highest accuracy) to ensure that the random forest model achieves efficient and accurate customer classification in inclusive credit scenarios.

[0089] The above embodiments introduce a customer classification method based on the industrial chain from the perspective of process flow. The following embodiments introduce a customer classification device based on the industrial chain from the perspective of virtual modules or virtual units. For details, please refer to the following embodiments.

[0090] This application provides a customer classification device based on the industry chain. Figure 4 A schematic diagram of a customer classification device based on the industrial chain is provided for an embodiment of this application, as shown below. Figure 4 As shown, the device includes:

[0091] Module 41 is used to acquire credit data of target customers to be classified; the credit data includes basic information data, loan behavior data and industry chain correlation data of target customers;

[0092] Extraction module 42 is used to extract features from the credit data of the target customer based on the chi-square test, and obtain the feature vector corresponding to the credit data;

[0093] Analysis module 43 is used to input feature vectors into a pre-built random forest model to obtain the classification results of target customers. The random forest model is optimized based on the improved spotted hyena algorithm. The improved spotted hyena algorithm optimizes the random forest model by co-optimizing at least one of the following: chaotic initialization of the population, adaptive fitness function, number of trees and number of leaf nodes.

[0094] In another possible implementation of the embodiments of the present application, the analysis module 43 is further configured to cause the improved spotted hyena algorithm to optimize the random forest model through the following process:

[0095] The population of the random forest model is initialized by using the chaotic mapping to generate a traversing sequence, to obtain an initialized population; each individual included in the initialized population is used to represent a combination of parameters, and the combination of parameters includes the number of trees and the number of leaf nodes;

[0096] The individual fitness of each combination of parameters in the initialized population is calculated by using a dynamic adaptive fitness function; the dynamic adaptive fitness function is constructed based on at least one of the following parameters: model classification accuracy, calculation overhead, and population parameter distance, so that an individual far away from the population mean has a higher selection probability;

[0097] Each combination of parameters is iteratively updated until a maximum iteration number is reached or the individual fitness value converges, and the optimal combination of parameters is output, to obtain the pre-constructed random forest model.

[0098] In another possible implementation of the embodiments of the present application, the analysis module 43 is specifically configured to:

[0099] Determine a search space of the number of trees and the number of leaf nodes;

[0100] A chaotic sequence is generated in a two-dimensional Euclidean space by using a Tent mapping function; the expression of the Tent mapping function is: wherein a is a control parameter, the value range of a is 0-1, d is used to represent the dimension, the value of d is 2, y i，d is used to represent the chaotic value of the ith individual in the dth dimension, and the value range of y

[0101] The chaotic sequence is mapped to the search space by using a linear transformation, to generate the initialized population.

[0102] In another possible implementation of the embodiments of the present application, the analysis module 43 is further configured to:

[0103] The population parameters are initialized, and the population parameters include the number of population individuals, the maximum iteration number, the initial individual fitness, and the individual position; wherein the initial individual fitness is calculated by using an internal 5-fold cross-validation strategy.

[0104] In another possible implementation of the embodiments of the present application, the analysis module 43 is specifically configured to:

[0105] The individual fitness of each individual in the current population is calculated based on a dynamic adaptive fitness function; the dynamic adaptive fitness function is expressed as wherein, a fitness value of the individual i in the t-1th iteration process, a fitness value corresponding to the optimal individual in the t-1th iteration process, a fitness value of the individual i in the tth iteration process, t is a current iteration number, and T is a maximum iteration number;

[0106] updating the individual position of at least part of individuals in the current population according to the updating formula to obtain an updated current population; the updating formula is represented as: wherein, an updated position of the individual i, a position of the individual i in the jth index corresponding search space in the last iteration process, j is used to represent the number of trees or the number of leaf nodes, a search upper bound and a search lower bound of a random number in the range of 0-1;

[0107] when the maximum iteration number is reached, and / or, the fitness value of the optimal individual in the continuous k iteration processes changes less than a preset threshold, the updated current population is taken as an optimal parameter combination, and a random forest model pre-constructed according to the optimal parameter combination is obtained.

[0108] In another possible implementation manner of the embodiment of the application, the random forest model pre-constructed is trained through the following process:

[0109] The original sample data are subjected to feature screening and parameter optimization based on the improved spotted hyena algorithm, to generate a preprocessed data set suitable for the random forest model;

[0110] The preprocessed data set is divided into k subsets with balanced category proportions by using stratified k-fold cross-validation, the random forest model is trained by iteratively using k-1 subsets and verifying the model performance by using one subset, the model training process is optimized through the comprehensive evaluation results after multiple iterations, and finally the random forest model is obtained.

[0111] In another possible implementation manner of the embodiment of the application, the extraction module 42 is specifically configured to:

[0112] A contingency table is constructed based on the category label and each data feature in the credit data, and actual observation values and theoretical values of each cell of the contingency table are calculated;

[0113] The actual observation values and the theoretical values are substituted into a chi-square test formula to calculate the chi-square values of the data features;

[0114] ​​​​According to the chi-square value in descending order, a feature importance list is obtained, and a target feature is filtered according to the feature importance list to generate a feature vector; the chi-square value of the target feature meets at least one of the following conditions: greater than a preset chi-square value, ranked greater than a preset rank.

[0115] In another possible implementation of the embodiment of the application, the extraction module 42 is specifically used for:

[0116] The target features filtered are subjected to standardization or normalization processing to obtain processed target features.

[0117] The processed target features are combined into a vector according to a selected feature order to obtain a feature vector.

[0118] In another possible implementation of the embodiment of the application, the device further includes a preprocessing module (not shown in the figure), which is used for:

[0119] The credit data is subjected to at least one of the following preprocessing: outlier processing, missing value processing, and data merging processing, to obtain processed credit data.

[0120] Correspondingly, the extraction module 42 is specifically used for:

[0121] The processed credit data is subjected to feature extraction in combination with chi-square test to obtain a feature vector corresponding to the credit data.

[0122] In another possible implementation of the embodiment of the application, the industry chain related data includes customer industry, customer asset size, customer associated enterprise information, industry proportion, upstream and downstream enterprise credit information, industry chain data, and industrial park data.

[0123] The embodiment of the application provides a customer classification device based on an industry chain, which is suitable for the method embodiment described above, and details are not repeated here.

[0124] In the embodiment of the application, an electronic device is provided, Figure 5 A structural schematic diagram of an electronic device provided by the embodiment of the application is shown in Figure 5 As shown in Figure 5 The electronic device shown in the figure includes a processor 51 and a memory 52. The processor 51 and the memory 52 are connected, such as through a bus 53. Optionally, the electronic device can further include a transceiver 54. It should be noted that in actual application, the transceiver 54 is not limited to one, and the structure of the electronic device does not constitute a limitation on the embodiment of the application.

[0125] The processor 51 can be a central processing unit (CPU), a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, transistor logic device, hardware component or any combination thereof. It can implement or execute various exemplary logical blocks, modules and circuits described in connection with the disclosure. The processor 51 can also be a combination of computing functions, such as one or more microprocessor combinations, combinations of DSP and microprocessor, etc.

[0126] The bus 53 can include a path that transmits information between the above-mentioned components. The bus 53 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 53 can be divided into an address bus, a data bus, a control bus, etc. For convenience of representation, Figure 5 Only one thick line is used in the figure, but it does not mean that there is only one bus 53 or one type of bus 53.

[0127] The memory 52 can be a read only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read only memory (EEPROM), a compact disc read only memory (CD-ROM) or other optical disk storage, an optical disk storage (including a compact disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but not limited to this.

[0128] The memory 52 is configured to store application program codes for implementing the solutions of the present application, and the processor 51 is configured to execute the application program codes stored in the memory 52.

[0129] The present application also provides a computer readable storage medium, which can include a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes. Specifically, the computer readable storage medium stores program instructions, and the program instructions are used to implement the method in each of the above embodiments.

[0130] The present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the technical solutions of the above method embodiments are implemented. The implementation principles and technical effects are similar, and thus will not be described here.

[0131] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed here. The present application is intended to cover any variations, uses or adaptive changes of the present application following the general principles of the present application and including common knowledge or conventional technical means in the art not disclosed in the present application. The specification and examples are only regarded as exemplary, and the true scope and spirit of the present application are indicated by the claims.

[0132] It should be understood that the present application is not limited to the precise construction that has been described above and shown in the accompanying drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present application is only limited by the appended claims.

Claims

1. An industry chain-based customer classification method, characterized in that, The method comprises: acquiring credit data of a target customer to be classified; the credit data comprises basic information data, loan behavior data and industry chain association data of the target customer; performing feature extraction on the credit data of the target customer based on a chi-square test to obtain a feature vector corresponding to the credit data; inputting the feature vector into a pre-constructed random forest model to obtain a classification result of the target customer; the random forest model is optimized based on an improved spotted hyena algorithm, and the improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of a population, adaptive fitness function, number of trees and number of leaf nodes in collaborative optimization.

2. The method of claim 1, wherein, The improved spotted hyena algorithm optimizes the random forest model through the following process: initializing a population of the random forest model using a chaotic mapping to generate a traversing sequence, to obtain an initialized population; each individual included in the initialized population is used to represent a set of parameter combinations, and the parameter combinations include the number of trees and the number of leaf nodes; calculating individual fitness of each parameter combination in the initialized population using a dynamically adaptive fitness function; the fitness function is constructed based on at least one of the following parameters: model classification accuracy, calculation overhead and population parameter distance, so that individuals far from the population mean have a higher selection probability; iteratively updating each parameter combination until a maximum iteration number is reached or the individual fitness value converges, and outputting an optimal parameter combination to obtain the pre-constructed random forest model.

3. The method of claim 2, wherein, The use of a chaotic mapping to generate a traversing sequence to initialize a population of the random forest model to obtain an initialized population comprises: determining a search space for the number of trees and the number of leaf nodes; A chaotic sequence is generated in a two-dimensional Euclidean space by a triangular mapping function Tent mapping function; an expression of the Tent mapping function is: Wherein a is a control parameter, the value range is 0-1, d is used to represent the dimension, the value is 2, is used to represent the number of the tree and the number of the leaf node, y i，d is used to represent the chaotic value of the ith individual in the dth dimension, and the value range is also 0-1. mapping the chaotic sequence to the search space through linear transformation to generate the initialized population.

4. The method according to claim 2 or 3, characterized in that, The method further comprises: initializing population parameters, including population individual number, maximum iteration number, initial individual fitness, and individual position; the initial individual fitness is calculated by an internal 5-fold cross-validation strategy.

5. The method according to claim 2 or 3, characterized in that, The iteratively updating each parameter combination until a maximum iteration number is reached or the individual fitness value converges, and outputting an optimal parameter combination to obtain the pre-constructed random forest model, comprises: calculating individual fitness of each individual in the current population based on the dynamic adaptive fitness function; the dynamic adaptive fitness function is expressed as wherein, is the individual fitness of individual i in the t-1th iteration process, is the individual fitness corresponding to the optimal individual in the t-1th iteration process, is the individual fitness of individual i in the tth iteration process, t is the current iteration number, and T is the maximum iteration number. updating individual positions of at least part of individuals in the current population according to an updating formula, to obtain an updated current population; the updating formula is expressed as: wherein, is the updated position of individual i, is the position of individual i in the search space corresponding to the jth index in the last iteration process, j is used to represent the number of trees or the number of leaf nodes, , is the search upper bound of , , , is a random number in the range of 0-1; when the maximum iteration number is reached, and / or the individual fitness of the optimal individual of continuous k iterations changes by less than a preset threshold, the updated current population is taken as the optimal parameter combination, and the pre-constructed random forest model is obtained according to the optimal parameter combination.

6. The method according to any one of claims 1 to 3, characterized in that, The pre-constructed random forest model is trained through the following process: performing feature screening and parameter optimization on original sample data based on an improved spotted hyena algorithm to generate a preprocessed data set adapted to the random forest model; The preprocessed data set is divided into k subsets with balanced category proportions by stratified k-fold cross-validation, and the random forest model is trained iteratively using k-1 subsets and validated using 1 subset. The model training process is optimized through comprehensive evaluation results after multiple iterations to obtain the final random forest model.

7. The method according to any one of claims 1 to 3, characterized in that, The feature extraction of the credit data of the target customer based on the chi-square test obtains a feature vector corresponding to the credit data, which includes: A contingency table is constructed based on the category label and each data feature in the credit data, and the actual observation value and the theoretical value of each cell of the contingency table are calculated; The actual observation value and the theoretical value are substituted into the chi-square test formula to calculate the chi-square value of each data feature; The data features are arranged in descending order of the chi-square value to obtain a feature importance list, and the target features are selected according to the feature importance list to generate the feature vector; the chi-square value of the target feature satisfies at least one of the following conditions: greater than a preset chi-square value, ranking greater than a preset ranking.

8. The method of claim 7, wherein, The feature extraction of the credit data of the target customer based on the chi-square test obtains a feature vector corresponding to the credit data, which includes: The target features obtained by screening are standardized or normalized to obtain processed target features; The processed target features are arranged in a selected feature order to form a vector to obtain the feature vector.

9. The method according to any one of claims 1 to 3, characterized in that, The method further includes: The credit data is preprocessed by at least one of the following: outlier processing, missing value processing, and data merging processing to obtain processed credit data; The feature extraction of the credit data of the target customer based on the chi-square test obtains a feature vector corresponding to the credit data, which includes: The feature extraction of the credit data of the target customer based on the chi-square test obtains a feature vector corresponding to the credit data, which includes:

10. The method according to any one of claims 1 to 3, characterized in that, The industry chain association data includes customer industry, customer asset size, customer associated enterprise information, industry proportion, upstream and downstream enterprise credit information, industry chain data, and industrial park data.

11. An industry chain-based customer classification device, characterized by comprising: The method includes: An acquisition module is configured to acquire credit data of a target customer to be classified, wherein the credit data includes basic information data, loan behavior data, and industry chain association data of the target customer; An extraction module is configured to extract features from the credit data of the target customer based on a chi-square test to obtain a feature vector corresponding to the credit data; An analysis module is configured to input the feature vector into a pre-constructed random forest model to obtain a classification result of the target customer; the random forest model is optimized based on an improved spotted hyena algorithm, and the improved spotted hyena algorithm optimizes the random forest model by at least one of the following: chaotic initialization of population, self-adaptive fitness function, and collaborative optimization of the number of trees and the number of leaf nodes.

12. An electronic device, comprising: It includes: At least one processor and a memory connected in communication with the processor; The memory stores computer execution instructions; The at least one processor executes the computer execution instructions stored in the memory to implement the method of any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method of any one of claims 1-10.

14. A computer program product, characterised in that, A computer program is included, which, when executed by a processor, implements the method of any one of claims 1-10.

Images