Credit object classification method and device based on Gaussian mixture distribution, equipment and medium

Abstract

The invention discloses a credit object classification method and device based on Gaussian mixture distribution, equipment and a medium, and relates to the field of financial science and technology, and the method comprises the steps: obtaining a credit key feature to determine a first Gaussian distribution component, and determining a target feature vector based on credit; determining an initial Gaussian probability density function according to the initial mean value, the initial covariance matrix and the target feature vector to construct an initial Gaussian mixture distribution model; obtaining a historical default rate index, determining a posterior probability based on the initial Gaussian mixture distribution model and the target feature vector to update the initial prior probability, the initial mean value and the initial covariance matrix, and obtaining a current likelihood function value to update the initial Gaussian mixture distribution model to obtain a target likelihood function value; determining a target Gaussian mixture distribution model based on the target likelihood function value; and identifying the second Gaussian distribution component by using the target Gaussian mixture distribution model so as to judge category features and / or risk levels and generate a credit management strategy. And reasonable distribution of credit resources is realized.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular to a credit object classification method, apparatus, device and medium based on Gaussian mixture distribution. Background Art

[0002] Fintech plays a vital role in the various businesses and operational management aspects of traditional banks, enabling digital transformation, intelligent upgrades, and modular decomposition across the entire process. Furthermore, as the scope of Fintech's application in the banking industry continues to expand, the use of artificial intelligence algorithms in this field is also becoming increasingly diverse. Credit risk control is a key component of risk management in banking processes, and AI algorithms, such as big data and machine learning models, are increasingly being applied to this process. Machine learning algorithms, in particular, are often used to categorize bank credit customers, facilitating loan application approval and interest rate adjustments by online systems or approval personnel.

[0003] With the continuous development of credit business, financial institutions face the challenge of accurately assessing and effectively managing numerous credit clients. Traditional credit client management methods often rely on relatively simple indicators or statistical models, which cannot fully and accurately depict the complex distribution of credit client characteristics. This leads to deficiencies in risk prediction, customer classification, and credit strategy formulation. For example, considering only a client's credit score may overlook the impact of comprehensive factors such as their actual capital expenditure and application approval rate on subsequent credit performance.

[0004] To sum up, how to achieve the rational allocation of credit resources is an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a credit object classification method, apparatus, device and medium based on Gaussian mixture distribution, which can achieve the rational allocation of credit resources. The specific scheme is as follows:

[0006] In a first aspect, the present application provides a credit object classification method based on Gaussian mixture distribution, comprising:

[0007] Acquire a key credit feature corresponding to a target object, determine a first Gaussian distribution component corresponding to the target object in a feature space based on the key credit feature, and determine a target feature vector of the target object based on the credit of the target object;

[0008] Determine an initial Gaussian probability density function according to the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component;

[0009] Obtaining a historical default rate indicator corresponding to the target object, determining a posterior probability of the first Gaussian distribution component based on an initial Gaussian mixture distribution model and the target eigenvector, and then updating the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain a current likelihood function value of the initial Gaussian mixture distribution model;

[0010] Iteratively updating the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determining a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model;

[0011] The target Gaussian mixture distribution model is used to identify the second Gaussian distribution component corresponding to the object to be classified, the category characteristics and / or risk level of the object to be classified are determined based on the second Gaussian distribution component, and the corresponding credit management strategy is generated based on the category characteristics and / or risk level of the object to be classified.

[0012] Optionally, determining an initial Gaussian probability density function according to an initial mean and an initial covariance matrix of the first Gaussian distribution component and the target eigenvector includes:

[0013] Determining the number of latitudes in the target feature vector; wherein the target feature vector includes any one or more of a pass rate, a spending rate, and a credit score;

[0014] An initial Gaussian probability density function is determined according to the number of latitudes in the target eigenvector, the initial mean and initial covariance matrix of the first Gaussian distribution component, and the target eigenvector.

[0015] Optionally, obtaining the key credit feature corresponding to the target object and determining the first Gaussian distribution component corresponding to the target object in the feature space based on the key credit feature includes:

[0016] Obtain key credit characteristics of target subjects in the target subject group;

[0017] Determining the number of Gaussian distribution components present in the target object group in the feature space;

[0018] A first Gaussian distribution component corresponding to the target object in the feature space is determined based on the key credit feature, so as to construct an initial Gaussian mixture distribution model based on the number of the Gaussian distribution components and the first Gaussian distribution component.

[0019] Optionally, updating the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator includes:

[0020] Superimposing the posterior probabilities of the first Gaussian components of the target objects in the target object group to obtain a sum of the posterior probabilities;

[0021] Obtaining a target ratio of the sum of the posterior probabilities to the number of the target subject group;

[0022] The initial prior probability is updated based on the target ratio and the historical default rate indicator using a maximum expectation algorithm to obtain an updated prior probability, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model according to the updated prior probability.

[0023] Optionally, updating the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator includes:

[0024] Performing a weighted average on the feature vectors of each target object in the target object group based on the posterior probability of the first Gaussian component to obtain a weighted feature vector;

[0025] An updated mean of the first Gaussian distribution component is determined based on the obtained weighted eigenvector and the historical default rate indicator, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model using the updated mean.

[0026] Optionally, updating the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator includes:

[0027] Determine the target difference between the characteristic vector of each target object and the updated mean value using a maximum expectation algorithm based on the obtained posterior probability and the historical default rate indicator;

[0028] A preset weighted summation process is performed on the obtained target difference value based on the posterior probability to obtain an updated covariance matrix, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model according to the updated covariance matrix.

[0029] Optionally, determining a target Gaussian mixture distribution model based on a target likelihood function value of the currently updated Gaussian mixture distribution model includes:

[0030] Determine whether the target likelihood function value of the currently updated Gaussian mixture distribution model is less than a preset likelihood function threshold;

[0031] If the target likelihood function value of the current updated Gaussian mixture distribution model is less than the preset likelihood function threshold, the current updated Gaussian mixture distribution model is determined to meet the preset model convergence condition, and the current updated Gaussian mixture distribution model is determined as the target Gaussian mixture distribution model based on the determination result.

[0032] In a second aspect, the present application provides a credit object classification device based on Gaussian mixture distribution, comprising:

[0033] a vector determination module, configured to obtain a credit key feature corresponding to a target object, determine a first Gaussian distribution component corresponding to the target object in a feature space based on the credit key feature, and determine a target feature vector of the target object based on the credit of the target object;

[0034] a model construction module, configured to determine an initial Gaussian probability density function based on the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component;

[0035] a parameter modification module, configured to obtain a historical default rate indicator corresponding to the target object, determine a posterior probability of the first Gaussian distribution component based on an initial Gaussian mixture distribution model and the target eigenvector, and then update the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain a current likelihood function value of the initial Gaussian mixture distribution model;

[0036] a model determination module, configured to iteratively update the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determine a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model;

[0037] a strategy generation module for identifying a second Gaussian distribution component corresponding to the object to be classified using the target Gaussian mixture distribution model, determining the category characteristics and / or risk level of the object to be classified based on the second Gaussian distribution component, and generating a corresponding credit management strategy based on the category characteristics and / or risk level of the object to be classified.

[0038] In a third aspect, the present application provides an electronic device, comprising:

[0039] Memory, used to store computer programs;

[0040] A processor is used to execute the computer program to implement the aforementioned credit object classification method based on Gaussian mixture distribution.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned credit object classification method based on Gaussian mixture distribution.

[0042] In summary, the present application first obtains the credit key features corresponding to the target object, determines the first Gaussian distribution component corresponding to the target object in the feature space based on the credit key features, and determines the target feature vector of the target object based on the credit of the target object; determines the initial Gaussian probability density function according to the initial mean and initial covariance matrix of the first Gaussian distribution component and the target feature vector, and constructs an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target feature vector, and the initial prior probability of the first Gaussian distribution component; obtains the historical default rate indicator corresponding to the target object, determines the posterior probability of the first Gaussian distribution component based on the initial Gaussian mixture distribution model and the target feature vector, and then determines the target object based on the obtained posterior probability, the target feature vector, and the target feature vector. The historical default rate indicator is used to update the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model to obtain the current likelihood function value of the initial Gaussian mixture distribution model; the initial Gaussian mixture distribution model is iteratively updated according to the current likelihood function value to obtain the target likelihood function value of the current updated Gaussian mixture distribution model, and the target Gaussian mixture distribution model is determined based on the target likelihood function value of the current updated Gaussian mixture distribution model; the target Gaussian mixture distribution model is used to identify the second Gaussian distribution component corresponding to the object to be classified, and the category characteristics and / or risk level of the object to be classified are determined according to the second Gaussian distribution component, and the corresponding credit management strategy is generated according to the category characteristics and / or risk level of the object to be classified. As can be seen from the above, the present application first obtains the key credit characteristics of the target object, determines its corresponding first Gaussian distribution component according to the distribution in the feature space, and constructs the initial Gaussian probability density function in combination with the target feature vector and the initial mean and covariance matrix of the component to form the initial Gaussian mixture distribution model. Then, the historical default rate indicator is introduced. By calculating the posterior probability of the first Gaussian distribution component, the initial prior probability, initial mean and initial covariance matrix are updated by combining the target eigenvector and the default data. The current likelihood function value is converged through iterative optimization, and the target Gaussian mixture distribution model is finally obtained. Finally, the target Gaussian mixture distribution model is used to identify the second Gaussian distribution component to which the object to be classified belongs. According to the second Gaussian distribution component, the category characteristics and / or risk level of the object to be classified are determined, and the corresponding credit management strategy is generated.In this way, this application comprehensively considers the three key characteristics of pass rate, expenditure rate and credit score, avoiding the one-sided evaluation results caused by single-factor considerations. At the same time, the Gaussian mixture distribution model has a strong fitting ability and can more accurately reflect the true distribution pattern of the customer group in the feature space. It also introduces the default rate as a basis for correction, making full use of the most critical risk feedback information in the credit business, ensuring that the accuracy and effectiveness of the model can always match the actual credit risk situation. The classification of review objects can achieve a reasonable allocation of credit resources, avoid wasting resources on high-risk and unstable-income customers, and improve the efficiency of resource allocation for the entire credit business. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0044] Figure 1 This is a flow chart of a credit object classification method based on Gaussian mixture distribution disclosed in this application;

[0045] Figure 2 A schematic diagram of a specific historical default rate Gaussian distribution disclosed in this application;

[0046] Figure 3 This is a flowchart of a specific credit object classification method based on Gaussian mixture distribution disclosed in this application;

[0047] Figure 4 This is a schematic diagram of the structure of a credit object classification device based on Gaussian mixture distribution disclosed in this application;

[0048] Figure 5 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] At present, with the continuous development of credit business, financial institutions are facing the challenge of accurately evaluating and effectively managing a large number of credit customers. Traditional credit customer management methods often rely on relatively single indicators or simple statistical models, which makes it difficult to fully and accurately characterize the complex characteristic distribution of credit customers, resulting in deficiencies in risk prediction, customer classification, and credit strategy formulation. For example, only considering the customer's credit score may ignore the impact of comprehensive factors such as their actual capital expenditure and application approval rate on subsequent credit performance. In order to solve the above technical problems, the present application discloses a credit object classification method, device, equipment and medium based on Gaussian mixture distribution, which can achieve a reasonable allocation of credit resources.

[0051] See also Figure 1 As shown, the embodiment of the present invention discloses a credit object classification method based on Gaussian mixture distribution, which may include:

[0052] Step S11: Acquire the key credit features corresponding to the target object, determine the first Gaussian distribution component corresponding to the target object in the feature space based on the key credit features, and determine the target feature vector of the target object based on the credit of the target object.

[0053] In this embodiment, the key credit features of the target objects within the target object group are first obtained; the number of Gaussian distribution components present in the feature space of the target object group is determined; and based on the key credit features, the first Gaussian distribution component corresponding to the target object in the feature space is determined, so that an initial Gaussian mixture distribution model is constructed based on the number of Gaussian distribution components and the first Gaussian distribution component. Specifically, the key credit features used to evaluate credit customers are determined, namely, the approval rate, the expenditure rate, and the customer credit score. For example, the approval rate may include the rate of customer credit application approvals, primarily referring to the initial review approval rate, the final review approval rate, and the expenditure approval rate; the expenditure rate may include the ratio of the customer's actual expenditure amount to the credit limit. Based on these key credit features, it can be assumed that the distribution of customers in the feature space follows GMM (Gaussian Mixture Distribution), and the first Gaussian distribution component corresponding to the target object in the feature space is determined.

[0054] It can be understood that the characteristic distribution of credit customer groups is often not a simple regular distribution, but rather more complex. The Gaussian mixture distribution is composed of multiple Gaussian distributions and has strong fitting capabilities. It can well adapt to the complex distribution forms of credit customers in multiple feature dimensions. Compared with simple linear models or single distribution models, it more accurately reflects the true distribution pattern of customer groups in the feature space, making the classification and risk assessment of customers more in line with the actual situation.

[0055] Furthermore, the credit customer group can be regarded as a mixture of multiple subgroups with different characteristics and risk attributes. The Gaussian mixture distribution is distributed through different components. Assuming that there are K Gaussian distribution components in the customer group, for each target object i, obtain its corresponding feature vector (in Indicates the pass rate, Indicates the expenditure rate, Represents the customer credit score), so that after determining the first Gaussian distribution component, eigenvector, and the number K of Gaussian distribution components corresponding to the target object in the feature space, an initial Gaussian mixture distribution model is constructed.

[0056] Step S12: determine an initial Gaussian probability density function based on the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component.

[0057] In this embodiment, the initial mean and initial covariance matrix of the kth Gaussian component are first obtained to determine the number of latitudes in the target feature vector; wherein the target feature vector includes any one or more of the pass rate, expenditure rate, and credit score; the initial Gaussian probability density function is determined based on the number of latitudes in the target feature vector, the initial mean and initial covariance matrix of the first Gaussian distribution component, and the target feature vector. Specifically, the initial mean of the kth Gaussian component is obtained. and the initial covariance matrix , and determine the feature vector latitude d. For example, here d can be set to 3, corresponding to the three features of pass rate, expenditure rate and customer credit score, and then according to the initial mean , initial covariance matrix , eigenvector latitude d and eigenvector Determine the initial Gaussian probability density function :

[0058] ;

[0059] in, is the initial mean; is the initial covariance matrix; d is the eigenvector latitude; k is a Gaussian component; is the target feature vector; is the initial mean , initial covariance matrix The initial Gaussian probability density function of .

[0060] Further, after obtaining the initial Gaussian probability density function, the initial Gaussian probability density function is used , target feature vector , the initial prior probability of the first Gaussian distribution component Construct an initial Gaussian mixture distribution model:

[0061] ;

[0062] in, is the posterior probability that each customer i belongs to the kth Gaussian component; is the initial mean; is the initial covariance matrix; is the initial prior probability; d is the feature vector latitude; i is the customer; k is a Gaussian component; is the target feature vector; is the initial mean , initial covariance matrix The initial Gaussian probability density function of ; j is a natural number not greater than K; K is the number of Gaussian distribution components.

[0063] Step S13: Obtain the historical default rate indicator corresponding to the target object, determine the posterior probability of the first Gaussian distribution component based on the initial Gaussian mixture distribution model and the target eigenvector, and then update the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain the current likelihood function value of the initial Gaussian mixture distribution model.

[0064] In this embodiment, it is necessary to introduce the customer's historical default rate indicator and use the maximum expectation algorithm, namely the EM algorithm, to modify the parameters of the initial Gaussian mixture distribution model in the previous step. For each Gaussian component k, the customer default rate belonging to this component is calculated. , use the EM algorithm to correct the initial prior probability , initial mean and the initial covariance matrix First, based on the parameters of the current given Gaussian mixture distribution model (initial prior probability , initial mean and the initial covariance matrix ) and the collected target feature vector , calculate the posterior probability that each customer i belongs to the kth Gaussian component .

[0065] Next, after getting Then, based on these probabilities and the customer's characteristic data and default rate data , combined with the calculation of maximizing the distance between the default rate distributions of different customer groups, the parameters of the Gaussian mixture distribution are updated.

[0066] In a specific embodiment, the posterior probabilities of the first Gaussian components of each target object in the target object group are superimposed to obtain the sum of the posterior probabilities; a target ratio of the sum of the posterior probabilities to the number of the target object group is obtained; and the maximum expectation algorithm is used to update the initial prior probability based on the target ratio and the historical default rate indicator to obtain an updated prior probability, so as to obtain the current likelihood function value of the initial Gaussian mixture distribution model based on the updated prior probability. Specifically, the prior probability is determined The update formula is:

[0067] ;

[0068] Where N is the total number of credit customers; is the posterior probability; is the updated prior probability. The prior probability is updated based on the proportion of the sum of the posterior probabilities of each customer i belonging to the kth Gaussian component to the total number of customers, so that the prior probability is more consistent with the proportion of each component in the actual data. At the same time, considering the default rate The impact of the default rate of a Gaussian component can be assumed to also conform to the Gaussian distribution. If there are two customer groups, high-risk and low-risk, then the default rate can be obtained as follows: Figure 2 The default rate Gaussian distribution shown in Figure 1 is used. The optimization goal can then be set to widen the Gaussian distribution of default rates between different Gaussian components, so that the spatial distance between the average default rate of the left tail of the higher-risk customer group and the average default rate of the left tail of the lower-risk customer group, the average default rate of the right tail of the higher-risk customer group and the average default rate of the right tail of the lower-risk customer group, and the mean default rate of the higher-risk customer group and the mean default rate of the lower-risk customer group are as wide as possible. Considering that most samples are near the mean, the mean is given a higher weight in the distance calculation formula. The weighted calculation of the distance formula is as follows:

[0069] ;

[0070] in, represents the left-tail average default rate of high-risk customer groups, represents the average default rate of high-risk customers, represents the right-tail average default rate of high-risk customer groups, represents the left-tail average default rate of the low-risk customer group, represents the average default rate of low-risk customers, represents the right-tail average default rate of the low-risk customer group; The calculated distance.

[0071] For customers with higher risk, their prior probability can be appropriately reduced. ,Will Zoom range to 0< <1 adjustment coefficient a, let , in order to reduce the "weight" of customers corresponding to this high-risk component in the subsequent classification process.

[0072] In another specific embodiment, based on the posterior probability of the first Gaussian component, the feature vectors of each target object in the target object group are weighted averaged to obtain a weighted feature vector; based on the obtained weighted feature vector and the historical default rate indicator, the updated mean of the first Gaussian distribution component is determined, so as to obtain the current likelihood function value of the initial Gaussian mixture distribution model using the updated mean. Specifically, the mean is determined The update formula is:

[0073] ;

[0074] Where N is the total number of credit customers; is the target feature vector; is the posterior probability that each customer i belongs to the kth Gaussian component; is the updated mean.

[0075] Furthermore, the feature vector of each customer According to the posterior probability that it belongs to the kth Gaussian component Perform weighted averaging to obtain the updated mean. For high-risk customer groups, use the updated The same approach uses distance weights , combined with the default rate, if there are more defaulting customers in a Gaussian component, from the perspective of feature space, the feature vectors of defaulting customers will cause the mean to shift towards a direction that is more biased towards high-risk characteristics. Through the above-mentioned weighted average update process, on the basis of considering the overall customer probability distribution, the mean is adjusted to a position that better reflects the true feature center of the component, so as to better distinguish customer groups with different risk levels.

[0076] In a third specific implementation, based on the obtained posterior probability and the historical default rate indicator, a maximum expectation algorithm is used to determine the target difference between the characteristic vector of each target object and the updated mean; based on the posterior probability, a preset weighted summation process is performed on the obtained target difference to obtain an updated covariance matrix, so as to obtain the current likelihood function value of the initial Gaussian mixture distribution model based on the updated covariance matrix. Specifically, the covariance matrix is determined The update formula is:

[0077] ;

[0078] Where N is the total number of credit customers; is the updated covariance matrix; is the posterior probability that each customer i belongs to the kth Gaussian component; is the updated mean; is the target feature vector.

[0079] The covariance matrix is then updated by calculating the target difference between each customer's feature vector and the updated mean, and performing a weighted summation based on the posterior probability. This aims to reflect the degree of dispersion of customer features within the same Gaussian component and the changing correlations between them. For Gaussian components with higher default rates, their updated covariance matrices may exhibit greater dispersion. For example, defaulters may differ more significantly from normal customers in characteristics such as approval rate, spending rate, and credit score, thereby more accurately depicting the characteristic distribution of this high-risk component customer group. Next, after updating the initial prior probability, initial mean, and initial covariance matrix, the current likelihood function value of the initial Gaussian mixture distribution model after each update is calculated.

[0080] Step S14: iteratively update the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determine a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model.

[0081] In this embodiment, after obtaining the current likelihood function value, the initial Gaussian mixture distribution model is iteratively updated based on the current likelihood function value and pre-set hyperparameters. After the update is complete, the target likelihood function value of the updated Gaussian mixture distribution model is obtained. It should be noted that the likelihood function can be constructed based on the probability distribution of all customers, for example, in the form of a log-likelihood function.

[0082] Furthermore, it is determined whether the target likelihood function value of the current updated Gaussian mixture distribution model is less than the preset likelihood function threshold; if the target likelihood function value of the current updated Gaussian mixture distribution model is less than the preset likelihood function threshold, it is determined that the current updated Gaussian mixture distribution model meets the preset model convergence condition, and based on the determination result, the current updated Gaussian mixture distribution model is determined as the target Gaussian mixture distribution model. Specifically, when the change in the likelihood function value is less than the preset likelihood function threshold, or the number of iterations reaches the preset maximum number of iterations, such as 100 times, the model is considered to have converged, and the current updated Gaussian mixture distribution model is determined as the target Gaussian mixture distribution model. At the same time, the obtained 、 and These are the final parameters of the Gaussian mixture distribution after correction and optimization of the default rate. By using the EM algorithm combined with the default rate to correct the parameters of the Gaussian mixture distribution, the model can be continuously optimized and adjusted according to the actual default situation of credit customers, and more accurately reflect the distribution patterns of customers with different risk levels in terms of approval rate, expenditure rate, credit score and other characteristics.

[0083] Step S15: Use the target Gaussian mixture distribution model to identify the second Gaussian distribution component corresponding to the object to be classified, determine the category characteristics and / or risk level of the object to be classified based on the second Gaussian distribution component, and generate a corresponding credit management strategy based on the category characteristics and / or risk level of the object to be classified.

[0084] In this embodiment, after determining the target Gaussian mixture distribution model, the key credit features of the new object to be classified are obtained, that is, the credit application approval rate of the new object to be classified is obtained. , expenditure rate and credit score , and generate new feature vectors based on the obtained credit key features , use the trained target Gaussian mixture distribution model to calculate the new sample The probability of belonging to the kth Gaussian component:

[0085] ;

[0086] in, is the probability of the kth Gaussian component of the object to be classified; is the updated mean; is the updated covariance matrix; is the updated prior probability; is the feature vector of the object to be classified; k is the Gaussian component; j is a natural number not greater than K; K is the number of Gaussian distribution components.

[0087] Further, according to The Gaussian component k corresponding to the maximum value in is used to determine whether the object to be classified belongs to the type characteristics or risk level represented by this component, so as to carry out subsequent work according to the pre-established credit management strategy for customers with different category characteristics or risk levels.

[0088] As can be seen from the above, the embodiment of the present application first obtains the key credit features of the target object, determines the corresponding first Gaussian distribution component based on the distribution in the feature space, and constructs the initial Gaussian probability density function by combining the target feature vector and the initial mean and covariance matrix of the component to form an initial Gaussian mixture distribution model. Then, the historical default rate indicator is introduced, and by calculating the posterior probability of the first Gaussian distribution component, the initial prior probability, initial mean and initial covariance matrix are updated by combining the target feature vector and the default data. The current likelihood function value is converged through iterative optimization, and the target Gaussian mixture distribution model is finally obtained. Finally, the target Gaussian mixture distribution model is used to identify the second Gaussian distribution component to which the object to be classified belongs, and the category characteristics and / or risk level of the object to be classified are determined based on the second Gaussian distribution component, thereby generating a corresponding credit management strategy. In this way, the embodiment of the present application comprehensively considers the three key characteristics of pass rate, expenditure rate and credit score, avoiding the one-sided evaluation results caused by single-factor considerations. At the same time, the Gaussian mixture distribution model has a strong fitting ability, which can more accurately reflect the true distribution pattern of the customer group in the feature space, and introduces the default rate as a basis for correction, making full use of the most critical risk feedback information in the credit business, ensuring that the accuracy and effectiveness of the model can always match the actual credit risk situation. The classification of review objects can achieve a reasonable allocation of credit resources, avoid wasting resources on high-risk and unstable-return customers, and improve the efficiency of resource allocation of the entire credit business.

[0089] Based on the above embodiment, this application discloses a credit object classification method based on Gaussian mixture distribution, which can achieve the reasonable allocation of credit resources. Figure 3 The credit object classification method based on Gaussian mixture distribution is described in detail.

[0090] This application first determines the key credit features used to evaluate credit customers, namely the approval rate, expenditure rate and customer credit score. Based on these key credit features, the first Gaussian distribution component corresponding to the target object in the feature space is determined. Assuming that there are K Gaussian distribution components in the customer group, for each target object i, its corresponding feature vector is obtained. (in Indicates the pass rate, Indicates the expenditure rate, Represents the customer credit score), so that after determining the first Gaussian distribution component, eigenvector, and the number K of Gaussian distribution components corresponding to the target object in the feature space, an initial Gaussian mixture distribution model is constructed.

[0091] Secondly, get the initial mean of the kth Gaussian component and the initial covariance matrix , and determine the eigenvector latitude d, and then according to the initial mean , initial covariance matrix , eigenvector latitude d and eigenvector Determine the initial Gaussian probability density function After obtaining the initial Gaussian probability density function, use the initial Gaussian probability density function , target feature vector , the initial prior probability of the first Gaussian distribution component Construct an initial Gaussian mixture distribution model.

[0092] Next, based on the parameters of the currently given Gaussian mixture distribution model (initial prior probability , initial mean and the initial covariance matrix ) and the collected target feature vector , calculate the posterior probability that each customer i belongs to the kth Gaussian component Introduce the customer's historical default rate indicator to modify the parameters of the initial Gaussian mixture distribution model in the previous step and obtain the updated prior probability , updated mean And the updated covariance matrix After updating the initial prior probability, initial mean, and initial covariance matrix, the current likelihood function value of the initial Gaussian mixture distribution model after each update is calculated.

[0093] Then, after obtaining the current likelihood function value, the initial Gaussian mixture distribution model is iteratively updated according to the current likelihood function value based on pre-set hyperparameters. After the update is completed, the target likelihood function value of the current updated Gaussian mixture distribution model is obtained. When the change in the likelihood function value is less than a preset likelihood function threshold, or the number of iterations reaches a preset maximum number of iterations, such as 100, the model is considered to have converged, and the current updated Gaussian mixture distribution model is determined as the target Gaussian mixture distribution model.

[0094] Finally, after determining the target Gaussian mixture distribution model, the credit key features of the new object to be classified are obtained, and a new feature vector is generated based on the obtained credit key features. , use the trained target Gaussian mixture distribution model to calculate the new sample The probability of belonging to the kth Gaussian component ,according to The Gaussian component k corresponding to the maximum value in is used to determine whether the object to be classified belongs to the type characteristics or risk level represented by this component, so as to carry out subsequent work according to the pre-established credit management strategy for customers with different category characteristics or risk levels.

[0095] In this way, resources are avoided from being wasted on high-risk and unstable-return customers, and the efficiency of resource allocation in the entire credit business is improved.

[0096] See also Figure 4 As shown, the embodiment of the present invention discloses a credit object classification device based on Gaussian mixture distribution, comprising:

[0097] A vector determination module 11 is configured to obtain a credit key feature corresponding to a target object, determine a first Gaussian distribution component corresponding to the target object in a feature space based on the credit key feature, and determine a target feature vector of the target object based on the credit of the target object;

[0098] A model construction module 12 is used to determine an initial Gaussian probability density function based on the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and to construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component;

[0099] a parameter modification module 13, configured to obtain a historical default rate indicator corresponding to the target object, determine a posterior probability of the first Gaussian distribution component based on the initial Gaussian mixture distribution model and the target eigenvector, and then update the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain a current likelihood function value of the initial Gaussian mixture distribution model;

[0100] a model determination module 14, configured to iteratively update the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determine a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model;

[0101] The strategy generation module 15 is used to use the target Gaussian mixture distribution model to identify the second Gaussian distribution component corresponding to the object to be classified, determine the category characteristics and / or risk level of the object to be classified based on the second Gaussian distribution component, and generate a corresponding credit management strategy based on the category characteristics and / or risk level of the object to be classified.

[0102] As can be seen from the above, this application first obtains the key credit features of the target object, determines its corresponding first Gaussian distribution component based on the distribution in the feature space, and constructs an initial Gaussian probability density function based on the target feature vector and the initial mean and covariance matrix of the component to form an initial Gaussian mixture distribution model. Then, the historical default rate indicator is introduced, and by calculating the posterior probability of the first Gaussian distribution component, the initial prior probability, initial mean and initial covariance matrix are updated by combining the target feature vector and the default data. The current likelihood function value is converged through iterative optimization, and the target Gaussian mixture distribution model is finally obtained. Finally, the target Gaussian mixture distribution model is used to identify the second Gaussian distribution component to which the object to be classified belongs, and the category characteristics and / or risk level of the object to be classified are determined based on the second Gaussian distribution component, thereby generating a corresponding credit management strategy. In this way, this application comprehensively considers the three key characteristics of pass rate, expenditure rate and credit score, avoiding the one-sided evaluation results caused by single-factor considerations. At the same time, the Gaussian mixture distribution model has a strong fitting ability and can more accurately reflect the true distribution pattern of the customer group in the feature space. It also introduces the default rate as a basis for correction, making full use of the most critical risk feedback information in the credit business, ensuring that the accuracy and effectiveness of the model can always match the actual credit risk situation. The classification of review objects can achieve a reasonable allocation of credit resources, avoid wasting resources on high-risk and unstable-income customers, and improve the efficiency of resource allocation for the entire credit business.

[0103] In some specific implementations, the model building module 12 may specifically include:

[0104] A latitude number acquisition unit, configured to determine the number of latitudes in the target feature vector; wherein the target feature vector includes any one or more of a pass rate, a spending rate, and a credit score;

[0105] A function determination unit is used to determine an initial Gaussian probability density function based on the number of latitudes in the target feature vector, the initial mean and initial covariance matrix of the first Gaussian distribution component, and the target feature vector.

[0106] In some specific implementations, the parameter correction module 13 may specifically include:

[0107] A credit key feature acquisition unit, used to acquire the credit key features of a target object in a target object group;

[0108] a quantity determination unit, configured to determine the quantity of Gaussian distribution components existing in the target object group in the feature space;

[0109] The first Gaussian distribution component determination unit is used to determine the first Gaussian distribution component corresponding to the target object in the feature space based on the credit key feature, so as to construct an initial Gaussian mixture distribution model based on the number of the Gaussian distribution components and the first Gaussian distribution component.

[0110] In some specific implementations, the parameter correction module 13 may specifically include:

[0111] a posterior probability sum obtaining unit, configured to superimpose the posterior probabilities of the first Gaussian components of the target objects in the target object group to obtain a sum of the posterior probabilities;

[0112] a target ratio obtaining unit, configured to obtain a target ratio of the sum of the posterior probabilities to the number of the target object group;

[0113] The first current likelihood function value acquisition unit is used to update the initial prior probability based on the target ratio and the historical default rate indicator using a maximum expectation algorithm to obtain an updated prior probability, so as to obtain the current likelihood function value of the initial Gaussian mixture distribution model according to the updated prior probability.

[0114] In some specific implementations, the parameter correction module 13 may specifically include:

[0115] a weighted feature vector acquisition unit, configured to perform weighted averaging on the feature vectors of each target object in the target object group based on the posterior probability of the first Gaussian component to obtain a weighted feature vector;

[0116] A second current likelihood function value acquisition unit is used to determine the updated mean of the first Gaussian distribution component based on the obtained weighted eigenvector and the historical default rate indicator, so as to use the updated mean to obtain the current likelihood function value of the initial Gaussian mixture distribution model.

[0117] In some specific implementations, the parameter correction module 13 may specifically include:

[0118] a target difference acquisition unit, configured to determine a target difference between the characteristic vector of each target object and the updated mean value using a maximum expectation algorithm based on the obtained posterior probability and the historical default rate indicator;

[0119] The third current likelihood function value acquisition unit is used to perform a preset weighted summation process on the acquired target difference value based on the posterior probability to obtain an updated covariance matrix, so as to obtain the current likelihood function value of the initial Gaussian mixture distribution model according to the updated covariance matrix.

[0120] In some specific implementations, the model determination module 14 may specifically include:

[0121] A target likelihood function value judgment unit is used to judge whether the target likelihood function value of the currently updated Gaussian mixture distribution model is less than a preset likelihood function threshold;

[0122] The target Gaussian mixture distribution model determination unit is used to determine whether the current updated Gaussian mixture distribution model meets the preset model convergence condition if the target likelihood function value of the current updated Gaussian mixture distribution model is less than the preset likelihood function threshold, and determine the current updated Gaussian mixture distribution model as the target Gaussian mixture distribution model based on the determination result.

[0123] Furthermore, the embodiment of the present application also discloses an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram should not be considered as any limitation to the scope of application of the present application.

[0124] Figure 5 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. This electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps of the credit object identification method based on a Gaussian mixture distribution disclosed in any of the aforementioned embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0125] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device. The communication protocol it follows is any communication protocol that can be applied to the technical solution of this application and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world. Its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0126] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0127] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including computer programs capable of implementing the Gaussian mixture distribution-based credit object identification method disclosed in any of the aforementioned embodiments and executed by the electronic device 20, the computer program 222 may further include computer programs capable of implementing other specific tasks.

[0128] Furthermore, this application discloses a computer-readable storage medium for storing a computer program; when executed by a processor, the computer program implements the aforementioned Gaussian mixture distribution-based credit object identification method. The specific steps of this method can be found in the corresponding content disclosed in the aforementioned embodiments and will not be further elaborated here.

[0129] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from the other embodiments. Reference can be made to the descriptions of the identical or similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and the relevant parts can be referred to the descriptions of the methods.

[0130] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0131] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0132] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0133] The above is a detailed introduction to the technical solution provided by the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for those skilled in the art, according to the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A credit object classification method based on Gaussian mixture distribution, characterized in that: include: Acquire a key credit feature corresponding to a target object, determine a first Gaussian distribution component corresponding to the target object in a feature space based on the key credit feature, and determine a target feature vector of the target object based on the credit of the target object; Determine an initial Gaussian probability density function according to the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component; Obtaining a historical default rate indicator corresponding to the target object, determining a posterior probability of the first Gaussian distribution component based on an initial Gaussian mixture distribution model and the target eigenvector, and then updating the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain a current likelihood function value of the initial Gaussian mixture distribution model; Iteratively updating the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determining a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model; The target Gaussian mixture distribution model is used to identify the second Gaussian distribution component corresponding to the object to be classified, the category characteristics and / or risk level of the object to be classified are determined based on the second Gaussian distribution component, and the corresponding credit management strategy is generated based on the category characteristics and / or risk level of the object to be classified.

2. The credit object classification method based on Gaussian mixture distribution according to claim 1, characterized in that: The determining of an initial Gaussian probability density function according to the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector comprises: Determining the number of latitudes in the target feature vector; wherein the target feature vector includes any one or more of a pass rate, a spending rate, and a credit score; An initial Gaussian probability density function is determined according to the number of latitudes in the target eigenvector, the initial mean and initial covariance matrix of the first Gaussian distribution component, and the target eigenvector.

3. The credit object classification method based on Gaussian mixture distribution according to claim 1, characterized in that: The step of obtaining a key credit feature corresponding to the target object and determining a first Gaussian distribution component corresponding to the target object in a feature space based on the key credit feature includes: Obtain key credit characteristics of target subjects in the target subject group; Determining the number of Gaussian distribution components present in the target object group in the feature space; A first Gaussian distribution component corresponding to the target object in the feature space is determined based on the key credit feature, so as to construct an initial Gaussian mixture distribution model based on the number of the Gaussian distribution components and the first Gaussian distribution component.

4. The credit object classification method based on Gaussian mixture distribution according to claim 3, characterized in that: The updating of the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator comprises: Superimposing the posterior probabilities of the first Gaussian components of the target objects in the target object group to obtain a sum of the posterior probabilities; Obtaining a target ratio of the sum of the posterior probabilities to the number of the target subject group; The initial prior probability is updated based on the target ratio and the historical default rate indicator using a maximum expectation algorithm to obtain an updated prior probability, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model according to the updated prior probability.

5. The credit object classification method based on Gaussian mixture distribution according to claim 4, characterized in that: The updating of the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator comprises: Performing a weighted average on the feature vectors of each target object in the target object group based on the posterior probability of the first Gaussian component to obtain a weighted feature vector; An updated mean of the first Gaussian distribution component is determined based on the obtained weighted eigenvector and the historical default rate indicator, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model using the updated mean.

6. The credit object classification method based on Gaussian mixture distribution according to claim 5, characterized in that: The updating of the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model according to the obtained posterior probability, the target eigenvector, and the historical default rate indicator comprises: Determine the target difference between the characteristic vector of each target object and the updated mean value using a maximum expectation algorithm based on the obtained posterior probability and the historical default rate indicator; A preset weighted summation process is performed on the obtained target difference value based on the posterior probability to obtain an updated covariance matrix, so as to obtain a current likelihood function value of the initial Gaussian mixture distribution model according to the updated covariance matrix.

7. The credit object classification method based on Gaussian mixture distribution according to any one of claims 1 to 6, characterized in that: The determining of a target Gaussian mixture distribution model based on a target likelihood function value of the currently updated Gaussian mixture distribution model includes: Determine whether the target likelihood function value of the currently updated Gaussian mixture distribution model is less than a preset likelihood function threshold; If the target likelihood function value of the current updated Gaussian mixture distribution model is less than the preset likelihood function threshold, the current updated Gaussian mixture distribution model is determined to meet the preset model convergence condition, and the current updated Gaussian mixture distribution model is determined as the target Gaussian mixture distribution model based on the determination result.

8. A credit object classification device based on Gaussian mixture distribution, characterized in that: include: a vector determination module, configured to obtain a credit key feature corresponding to a target object, determine a first Gaussian distribution component corresponding to the target object in a feature space based on the credit key feature, and determine a target feature vector of the target object based on the credit of the target object; a model construction module, configured to determine an initial Gaussian probability density function based on the initial mean and initial covariance matrix of the first Gaussian distribution component and the target eigenvector, and construct an initial Gaussian mixture distribution model using the initial Gaussian probability density function, the target eigenvector, and the initial prior probability of the first Gaussian distribution component; a parameter modification module, configured to obtain a historical default rate indicator corresponding to the target object, determine a posterior probability of the first Gaussian distribution component based on an initial Gaussian mixture distribution model and the target eigenvector, and then update the initial prior probability, the initial mean, and the initial covariance matrix in the initial Gaussian mixture distribution model based on the obtained posterior probability, the target eigenvector, and the historical default rate indicator to obtain a current likelihood function value of the initial Gaussian mixture distribution model; a model determination module, configured to iteratively update the initial Gaussian mixture distribution model according to the current likelihood function value to obtain a target likelihood function value of the current updated Gaussian mixture distribution model, and determine a target Gaussian mixture distribution model based on the target likelihood function value of the current updated Gaussian mixture distribution model; a strategy generation module for identifying a second Gaussian distribution component corresponding to the object to be classified using the target Gaussian mixture distribution model, determining the category characteristics and / or risk level of the object to be classified based on the second Gaussian distribution component, and generating a corresponding credit management strategy based on the category characteristics and / or risk level of the object to be classified.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the credit object classification method based on Gaussian mixture distribution as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the credit object classification method based on Gaussian mixture distribution as described in any one of claims 1 to 7.

Images