End-to-end automated pipeline credit modeling method and system for inclusive finance scenarios

Abstract

The application discloses an end-to-end automatic pipeline credit modeling method and system for a general financial scenario, and relates to the technical field of automatic modeling. The method is to first acquire a large amount of sample data and perform automatic cleaning and variable type determination, then for continuous variables, adopt a decision tree algorithm to perform automatic binning processing, then based on the binning results or discrete conditions, automatically calculate the information value of each variable and the correlation coefficient between variables, and complete the automatic screening of variables according to a preset threshold, finally, import the remaining variables obtained through screening and the original dependent variables into a machine learning model for model training, obtain a loan default prediction model and output the same, so that the whole process from the original data to the deployed model can be automated, and the modeling efficiency, objectivity and reproducibility in the multi-dimensional and big data scenario of the general financial field are significantly improved, and the method is convenient for practical application and popularization.

Description

Technical Field

[0001] This invention belongs to the field of automated modeling technology, specifically relating to an end-to-end automated pipeline credit modeling method and system for inclusive finance scenarios, which is particularly suitable for full-process modeling of pre-loan approval risk control models in inclusive finance scenarios. Background Technology

[0002] With the deepening development of inclusive finance (which refers to providing appropriate and effective financial services to all social strata and groups with financial service needs at an affordable cost, based on the principles of equal opportunity and commercial sustainability), the coverage of financial services is constantly expanding, and the scale of credit business targeting micro and small enterprises and individuals continues to grow. In this type of business, pre-loan risk control is the core link, and its key lies in building accurate and efficient credit assessment models. Traditional credit risk modeling methods usually rely on the manual operation and experience judgment of statistical experts or data scientists, and their general process includes multiple discrete steps such as data cleaning, variable selection, and model training.

[0003] However, in inclusive finance scenarios, applicants (especially micro and small enterprises) often lack the strong financial attribute data (such as collateral and complete financial statements) valued by traditional financial institutions. Instead, they provide a large amount of multi-dimensional alternative data, such as the business owner's personal information, historical behavioral data, operating cash flow, and tax information. This dramatically increases the sample size (often exceeding thousands or even tens of thousands) and the number of variables (often exceeding hundreds) required for modeling. This results in characteristics such as large sample size, complex variable distribution, and difficulty in covering variables with manual rules, leading to increasingly prominent limitations of traditional modeling methods, mainly in the following aspects: (1) The process is not very automated and is time-consuming and labor-intensive. In other words, key steps such as variable screening in existing technologies rely heavily on manual intervention. For example, variable screening requires manual calculation and analysis of a large number of statistical indicators (such as correlation coefficients). In the context of a surge in sample and variable dimensions, this manual or semi-automatic operation mode requires a lot of time and human resources, resulting in long modeling cycles and low efficiency, which makes it difficult to meet the needs of rapid business iteration. (2) It is highly dependent on subjective experience and difficult to standardize. That is, many key modeling decisions depend on the personal experience of modeling experts. This experience-based technical path has strong subjectivity. Different experts may draw different conclusions, making it difficult to standardize and reproduce the modeling process and results. At the same time, artificial rules are difficult to exhaust the potential patterns hidden in complex data, which may lead to insufficient mining of the value of data information and affect the final performance of the model. (3) Knowledge solidification and deployment difficulties: because the above process relies on expert experience, its core decision-making logic is difficult to be completely and accurately abstracted and solidified into a unified modeling pipeline. This makes each modeling a "manual operation", and successful experience is difficult to be directly copied and promoted to new business scenarios or datasets. Therefore, existing technologies are difficult to support stable and repeatable one-click automated modeling deployment, which restricts the large-scale and efficient application of risk control models.

[0004] Therefore, there is an urgent need for a technical solution that can overcome the above-mentioned shortcomings and achieve end-to-end and fully automated processing from raw data to credit models in the specific scenario of inclusive finance, so as to minimize human intervention and improve modeling efficiency, objectivity and replicability. Summary of the Invention

[0005] The purpose of this invention is to provide an end-to-end automated pipeline credit modeling method, system, computer equipment, computer-readable storage medium, and computer program product for inclusive finance scenarios, in order to solve the problems of low process automation, strong reliance on subjective experience, and / or difficulty in knowledge solidification and deployment in existing credit risk modeling schemes for inclusive finance scenarios.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, an end-to-end automated pipeline credit modeling method for inclusive finance scenarios is provided, including: Obtain M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data contains N independent variables as model inputs and one dependent variable as model outputs, where N represents a positive integer greater than 100. The N independent variables reflect multi-dimensional information about the lender before the loan is issued, and the dependent variable indicates whether the lender defaults after the loan is issued. For each sample data in the M sample data, data cleaning processing is performed on the N independent variables in the corresponding data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes text-to-numerical processing and missing value processing. For each variable among the N cleaned independent variables, determine whether the corresponding variable is a discrete variable or a continuous variable based on all the observed values ​​of the corresponding variable, wherein the all observed values ​​include M observed values ​​that correspond one-to-one with the M sample data. For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, the decision tree algorithm is used to automatically bin the corresponding variable to obtain the binning result of the corresponding variable. For each variable, based on the binning results or discreteness of the corresponding variable and the dependent variable in the M sample data, the information value of the corresponding variable is calculated, wherein the information value is used to positively measure the predictive ability of the corresponding variable for loan default. For each pair of variables among all the cleaned independent variables, the correlation coefficient between the two variables is calculated based on the binning or discreteness of the two variables. When the absolute value of the correlation coefficient is greater than or equal to a first preset threshold, the variable with a smaller information value among the two variables is removed. The remaining cleaned independent variables and dependent variables from the M sample data are imported into a machine learning model for model training to obtain a loan default prediction model, which is then output.

[0007] Based on the above-mentioned invention, a new solution is provided that enables end-to-end and fully automated processing from raw data to credit models in the specific scenario of inclusive finance. This involves first acquiring a large amount of sample data and performing automated cleaning and variable type determination. Then, for continuous variables, a decision tree algorithm is used for automated binning. Next, based on the binning results or discrete cases, the information value of each variable and the correlation coefficient between variables are automatically calculated, and variables are automatically filtered according to preset thresholds. Finally, the remaining variables obtained from the filtering, along with the original dependent variable, are imported into a machine learning model for model training, resulting in a loan default prediction model, which is then output. This achieves full automation from raw data to model deployment, significantly improving the modeling efficiency, objectivity, and reproducibility in multi-dimensional and big data scenarios of inclusive finance, facilitating practical application and promotion.

[0008] In one possible design, for each variable among the N cleaned independent variables, based on all the observed values ​​of the corresponding variable, it is determined whether the corresponding variable is a discrete variable or a continuous variable, including: For each variable among the N cleaned independent variables, based on all the observed values ​​of the corresponding variable, the number of unique values ​​of the corresponding variable is counted. Then, it is determined whether the number of unique values ​​is less than or equal to a second preset threshold. If so, the corresponding variable is determined to be a discrete variable; otherwise, the corresponding variable is determined to be a continuous variable. Here, all the observed values ​​include M observed values ​​that correspond one-to-one with the M sample data. The number of unique values ​​refers to the total number of non-repeating values ​​obtained after deduplication of all the observed values ​​of the corresponding variable.

[0009] In one possible design, for any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, a decision tree algorithm is used to automatically bin the corresponding variable, obtaining the binning result of the corresponding variable, including: For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, then all observations of the corresponding variable are obtained, wherein the all observations include M observations that correspond one-to-one with the M sample data. By applying all the observed values ​​of any variable and the dependent variable in the M sample data, and with the AUC value of the decision tree model as the optimization objective, the multiple hyperparameters of the decision tree algorithm are optimized to obtain the optimal values ​​of the multiple hyperparameters, wherein the multiple hyperparameters include the minimum number of sample leaf nodes, the complexity pruning parameter and / or the maximum branch depth. The optimal values ​​of the multiple hyperparameters are substituted into the decision tree algorithm to obtain the optimal decision tree algorithm. The optimal decision tree algorithm is then applied to bin the observations of any variable to obtain multiple bins of the observations of any variable. Based on the comparison results of the sample proportion and the third preset threshold, the bins of multiple observations of any variable are merged to obtain the final binning result of any variable where the sample proportion of each bin is greater than or equal to the third preset threshold. The sample proportion refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M.

[0010] In one possible design, based on the comparison result of the sample proportion and the third preset threshold, the multiple observations of any variable are binned and merged to obtain the final binning result where the sample proportion of any variable in each bin is greater than or equal to the third preset threshold, including the following steps S441 to S442: S441. Traverse each observation bin of any variable as follows: Calculate the sample percentage of the currently traversed bin. If the sample percentage is less than a third preset threshold, first calculate the first sample default rate of the currently traversed bin and the second sample default rate of at least one observation bin adjacent to the currently traversed bin. Then merge the currently traversed bin with an observation bin in the at least one observation bin that has the second sample default rate closest to the first sample default rate to obtain a new observation bin for any variable. Then traverse the next observation bin. Otherwise, directly traverse the next observation bin. Finally, after traversal is completed, execute step S442. Here, the sample percentage refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M. The first sample default rate and the second sample default rate refer to the ratio of the number of defaulting samples to the sample size corresponding to the corresponding observation bin, respectively. S442. Determine whether the sample proportion of each observation bin of any variable is greater than or equal to the third preset threshold. If so, take each observation bin of any variable as the final binning result; otherwise, return to step S441.

[0011] In one possible design, for each variable, based on the binning result or discreteness of the corresponding variable and the dependent variable in the M sample data, the information value of the corresponding variable is calculated, including: For any variable among the N cleaned independent variables, if the corresponding variable was previously determined to be a continuous variable, then obtain the corresponding binning result. For each observation in the binning results of a certain variable, the corresponding evidence weight value is calculated according to the following formula based on all corresponding sample data and the dependent variable in the M sample data:

[0012] In the formula, Represents positive integers. This indicates the first binning result of a certain variable. The evidence weight value for each bin of observations This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of default samples in all sample data corresponding to each observation bin. This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of non-default samples in all sample data corresponding to each observation bin. This represents the total number of default samples determined based on the dependent variable from the M sample data. This represents the total number of non-default samples determined based on the dependent variable in the M sample data; Based on the evidence weight values ​​of each bin for the observed values, the information value of a certain variable is calculated according to the following formula. :

[0013] In the formula, This represents the total number of bins in the binning results of a certain variable, and the information value is used to positively measure the predictive power of the corresponding variable for post-loan default.

[0014] In one possible design, after calculating the information values ​​of each variable and before calculating the correlation coefficient between the two variables, the method further includes: Remove variables whose information values ​​are less than a fourth preset threshold from the N cleaned independent variables.

[0015] Secondly, an end-to-end automated pipeline credit modeling system for inclusive finance scenarios is provided, including a sample data acquisition unit, a data cleaning and processing unit, a variable type judgment unit, a decision binning processing unit, an information value calculation unit, a variable correlation screening unit, and a prediction model training unit. The sample data acquisition unit is used to acquire M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data contains N independent variables used as model inputs and one dependent variable used as model outputs, where N represents a positive integer greater than 100. The N independent variables are used to reflect the multi-dimensional information of the lender before the loan is issued, and the dependent variable is used to indicate whether the lender has defaulted after the loan is issued. The data cleaning and processing unit is communicatively connected to the sample data acquisition unit. It is used to perform data cleaning processing on the N independent variables in the corresponding data for each of the M sample data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes text-to-numerical processing and missing value processing. The variable type determination unit is communicatively connected to the data cleaning and processing unit. It is used to determine whether a variable is a discrete variable or a continuous variable for each variable among the N cleaned independent variables based on all the observations of the corresponding variable. The observations include M observations that correspond one-to-one with the M sample data. The decision binning processing unit is communicatively connected to the variable type determination unit. It is used to automatically bin the corresponding variable using a decision tree algorithm if any variable among the N cleaned independent variables is determined to be a continuous variable, and to obtain the binning processing result of the corresponding variable. The information value calculation unit is communicatively connected to the data cleaning and processing unit and the decision binning and processing unit, respectively. It is used to calculate the information value of each variable based on the binning and processing result or discreteness of the corresponding variable and the dependent variable in the M sample data. The information value is used to positively measure the predictive ability of the corresponding variable for loan default. The variable correlation screening unit is communicatively connected to the data cleaning and processing unit, the decision binning and processing unit, and the information value calculation unit. It is used to calculate the correlation coefficient between the two variables for each pair of variables in all the cleaned independent variables, based on the binning and processing results or the discreteness of the two variables. When the absolute value of the correlation coefficient is greater than or equal to a first preset threshold, the variable with a smaller information value in the two variables is removed. The prediction model training unit is communicatively connected to the variable correlation screening unit, and is used to import all the remaining cleaned independent variables and dependent variables from the M sample data into the machine learning model for model training, so as to obtain the loan default prediction model and output it.

[0016] Thirdly, the present invention provides a computer device comprising a storage module, a processing module, and a transceiver module connected in sequence for communication, wherein the storage module is used to store a computer program, the transceiver module is used to send and receive messages, and the processing module is used to read the computer program and execute the end-to-end automated pipeline credit modeling method as described in the first aspect or any possible design in the first aspect.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the end-to-end automated pipeline credit modeling method as described in the first aspect or any possible design within the first aspect.

[0018] Fifthly, the present invention provides a computer program product, including a computer program or instructions, which, when executed by a computer, implement the end-to-end automated pipeline credit modeling method as described in the first aspect or any possible design in the first aspect.

[0019] The beneficial effects of the above scheme are: (1) This invention creatively provides a new solution that can realize end-to-end and full-process automated processing from raw data to credit model in the specific scenario of inclusive finance. That is, firstly, a large amount of sample data is acquired and automatically cleaned and the variable type is determined. Then, for continuous variables, the decision tree algorithm is used for automated binning. Then, based on the binning results or discrete cases, the information value of each variable and the correlation coefficient between variables are automatically calculated. The variables are automatically screened according to the preset threshold. Finally, the remaining variables obtained by screening and the original dependent variables are imported into the machine learning model for model training to obtain the loan default prediction model and output it. In this way, the full automation from raw data to model deployment can be realized, which significantly improves the modeling efficiency, objectivity and reproducibility in the multi-dimensional and big data scenarios of inclusive finance. (2) It can realize true end-to-end fully automated modeling, which greatly improves efficiency. That is, from data acquisition, cleaning, variable determination, binning, filtering to final model training, all links are integrated into an automated pipeline that must be executed in sequence. This directly solves the fundamental problems of "time-consuming and labor-intensive" and "difficult to automate deployment" caused by the reliance on manual operation and the fragmentation of each link in the traditional modeling method. It enables the entire process to be executed with one click through code, significantly shortening the modeling cycle and providing the ability to standardize deployment and reproduce. (3) By replacing subjective experience-based decision-making with objective quantitative rules, the standardization of the process and the stability of the results are ensured. That is, a clear threshold based on the "number of unique values" is introduced to determine the variable type, eliminating the subjective arbitrariness of expert experience. This also constitutes the core of automation in the binning process: not only does it automatically optimize the binning hyperparameters with the model AUC value as an objective goal to ensure the best binning effect, but it also innovatively adopts the dual condition rule of "sample proportion" and "default rate proximity" to perform post-processing merging of binning. This ensures that each bin has both statistical reliability and risk monotonicity in a business sense, so that the final generated model has both excellent predictive performance and good business interpretability. (4) A two-stage automated screening strategy is adopted to deeply mine data information to optimize model performance. Based on the unified results generated by the binning mentioned above, information values ​​are automatically calculated and variables with weak predictive ability are removed. Then, redundant information is automatically removed from the remaining variables based on the correlation coefficient. This sequential design of first screening information value and then screening correlation can systematically complete the elimination of variables without human intervention, maximize the retention of effective information and reduce multicollinearity, thus laying a solid foundation for training a predictive model with better risk control performance. (5) This solution transforms the traditional modeling process, which relies on subjective experience of experts and is discrete and inefficient, into an objective, continuous and automated high-efficiency pipeline. This not only achieves the primary goals of "improving efficiency" and "facilitating deployment", but also achieves the higher-order goals of "enhancing model performance" and "ensuring model stability and interpretability" through a series of ingenious algorithm processing designs. This produces a synergistic technical effect that is better than the simple sum of the parts, making it easier for practical application and promotion. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating an end-to-end automated pipeline credit modeling method for inclusive finance scenarios provided in an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the structure of an end-to-end automated pipeline credit modeling system for inclusive finance scenarios provided in an embodiment of this application.

[0023] Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these embodiments without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0025] It should be understood that although the terms "first" and "second", etc., may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object may be referred to as the second object, and similarly, the second object may be referred to as the first object, without departing from the scope of the exemplary embodiments of the invention.

[0026] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, or A and B exist simultaneously. Another example is A, B and / or C, which can mean that any one of A, B, and C or any combination thereof exists. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone or A and B exist simultaneously. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0027] Example like Figure 1As shown, the end-to-end automated pipeline credit modeling method for inclusive finance scenarios provided in the first aspect of this embodiment can be executed, but is not limited to, by computer devices with certain computing resources, such as servers, personal computers (PCs, referring to a type of multi-purpose computer of a size, price, and performance suitable for personal use; desktops, laptops, mini-laptops, tablets, and ultrabooks are all considered personal computers), smartphones, personal digital assistants (PDAs), or wearable devices. Figure 1 As shown, the end-to-end automated pipeline credit modeling method includes, but is not limited to, the following steps S1 to S7.

[0028] S1. Obtain M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data includes, but is not limited to, N independent variables used as model inputs and one dependent variable used as model outputs, where N represents a positive integer greater than 100. The N independent variables are used to reflect multi-dimensional information about the lender before the loan is issued, and the dependent variable is used to indicate whether the lender defaults after the loan is issued.

[0029] In step S1, the M sample data represent the original observation data for all historical borrowers. The multi-dimensional information includes, but is not limited to, the business owner's gender, age, education level, number of historical loans, outstanding loan balance, number of historical delinquencies, registered capital, years of operation, VAT payable, and / or income tax payable. The dependent variable can specifically use a value of "0" to represent no default and a value of "1" to represent default. Furthermore, the sample data can be obtained through conventional collection and transmission from existing lending platforms.

[0030] S2. For each sample data in the M sample data, perform data cleaning processing on the N independent variables in the corresponding data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes, but is not limited to, text-to-numerical processing and missing value processing.

[0031] In step S2, the N cleaned independent variables correspond one-to-one with the N original independent variables. Since the specific observations of the independent variables in the corresponding sample data may be descriptive text, and subsequent data processing and modeling require the independent variables to be digitized, the text-to-numerical conversion process is necessary. For example, gender (male) is converted to the numerical value "1", and gender (female) is converted to the numerical value "2". Furthermore, the specific method for handling missing values ​​can be exemplified by automatically converting missing values ​​to the special numerical value "-9999".

[0032] S3. For each variable among the N cleaned independent variables, determine whether the corresponding variable is a discrete variable or a continuous variable based on all the observed values ​​of the corresponding variable. The observed values ​​include, but are not limited to, M observed values ​​that correspond one-to-one with the M sample data.

[0033] In step S3, the observed value is the numerical value obtained after the data cleaning process. It can be the original observed value, the value converted from the original observed text, or a special value converted due to missing original observed content. Consider that the number of unique values ​​for different variables (which refers to the total number of unique values ​​obtained after deduplication of all observed values ​​for the corresponding variable) will vary; for example, for the variable gender, there are generally only 2 values ​​(i.e., the value "1" and the value "2", with a corresponding number of unique values ​​of 2), while for the variable age, there may be more than 60 values ​​(i.e., a corresponding number of unique values ​​of 60+). For the latter, in order to effectively improve the stability and interpretability of the subsequent model, it is necessary to bin the observed values ​​(i.e., divide them into a small number of bins with significantly different risk levels, such as dividing them into multiple intervals such as [18,25], (25,35], (35,45], (45,55], and (55,65]). Specifically, for each variable among the N cleaned independent variables, according to the corresponding strain... The method involves considering all observed values ​​of a quantity and determining whether the corresponding variable is discrete or continuous. This includes, but is not limited to, counting the number of unique values ​​for each variable among the N cleaned independent variables, based on all observed values. Then, it is determined whether the number of unique values ​​is less than or equal to a second preset threshold. If so, the corresponding variable is considered discrete; otherwise, it is considered continuous. The "all observed values" include, but are not limited to, M observed values ​​corresponding one-to-one with the M sample data. The number of unique values ​​refers to the total number of non-repeating values ​​obtained after deduplication of all observed values ​​for the corresponding variable. Based on the definition of the number of unique values, it can be conventionally calculated and is less than or equal to M. Furthermore, the second preset threshold can be, for example, but is not limited to, 10.

[0034] S4. For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, the decision tree algorithm is used to automatically bin the corresponding variable to obtain the binning result of the corresponding variable.

[0035] In step S4, the decision tree algorithm is an existing algorithm for dividing continuous variables into intervals based on the bisection rule of minimum purity invariance, and is commonly used in binning processes. To achieve optimal binning, it is necessary to introduce a dependent variable as guiding information during the binning process, and then use the information gain between the dependent and independent variables as a metric for measuring the binning effect. Preferably, for any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, the decision tree algorithm is used to automatically bin the corresponding variable, obtaining the binning result for the corresponding variable, including but not limited to the following steps S41 to S44.

[0036] S41. For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, then obtain all the observations of the corresponding variable, wherein the all the observations include, but are not limited to, M observations that correspond one-to-one with the M sample data.

[0037] S42. Applying all the observed values ​​of any variable and the dependent variable in the M sample data, and with the AUC value of the decision tree model as the optimization objective, optimize multiple hyperparameters of the decision tree algorithm to obtain the optimal values ​​of the multiple hyperparameters, wherein the multiple hyperparameters include the minimum number of sample leaf nodes, complexity pruning parameters and / or the maximum branch depth.

[0038] In step S42, since the M observations correspond one-to-one with the M sample data, each observation corresponds to a specific value of the dependent variable (i.e., "0" or "1"). This correspondence can then be used to form information gain, ultimately affecting the AUC (Area Under the Curve) value of the decision tree model. This is equivalent to using the information gain between the dependent and independent variables as a measure of binning effectiveness. In the specific optimization process, it can be implemented as follows: First, set the initial values ​​of the minimum number of sample leaf nodes, the complexity pruning parameter, and the maximum branch depth (generally set to the minimum value; for example, setting the initial value of the minimum number of sample leaf nodes to 100, the initial value of the complexity pruning parameter to 0.001, and the initial value of the maximum branch depth to 2, respectively). Also, set the step size for each parameter (for example, setting the step size of the minimum number of sample leaf nodes to 100, the step size of the complexity pruning parameter to 0.001, and the step size of the maximum branch depth to 1, respectively) and the maximum value (for example, the step size of the minimum number of sample leaf nodes to 100, the step size of the complexity pruning parameter to 0.001, and the step size of the maximum branch depth to 1, respectively). The maximum value of the minimum number of leaf nodes is set to 10000, the maximum value of the complexity pruning parameter is 0.05, and the maximum value of the maximum branch depth is 5. Then, starting from the initial value of each parameter according to the step size, the decision tree algorithm is traversed and iterated using all observed values ​​of any variable and the dependent variable in the M sample data. When the AUC value reaches the optimal (i.e., the maximum value), the parameter is at its optimal value. Furthermore, optimization algorithms such as particle swarm optimization, genetic optimization, or gray wolf optimization can be combined to complete the optimization process, so as to quickly obtain the optimal values ​​of the multiple hyperparameters. In addition, the specific application of all observed values ​​of any variable and the dependent variable in the M sample data in the optimization process can be derived conventionally by referring to the existing binning process based on decision tree algorithms, and will not be elaborated here.

[0039] S43. Substitute the optimal values ​​of the multiple hyperparameters into the decision tree algorithm to obtain the optimal decision tree algorithm, and apply the optimal decision tree algorithm to bin the observations of any variable to obtain multiple bins of the observations of any variable.

[0040] In step S43, the observation bin is a numerical range; for example, when any variable is age, the multiple observation bins may be [18,25], (25,35], (35,45], (45,55], and (55,65], etc. Each observation bin contains at least one non-repeating value of any variable and corresponds to at least one sample data; for example, when any variable is age, an observation bin of (25,35] may contain two non-repeating values, 27 and 32 years old. 27 years old may correspond to 3 sample data (i.e., the age is 27 in the three sample data), and 32 years old may correspond to 2 sample data (i.e., the age is 32 in the other two sample data). Therefore, an observation bin contains two non-repeating values ​​and corresponds to a total of 5 sample data.

[0041] S44. Based on the comparison result of the sample proportion and the third preset threshold, the multiple observations of any variable are binned and merged to obtain the final binning result of any variable where the sample proportion of each bin is greater than or equal to the third preset threshold. The sample proportion refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M.

[0042] In step S44, referring to the example in step S43, since a certain observation bin corresponds to a total of 5 sample data, that is, the sample size corresponding to a certain observation bin is 5, the sample proportion of a certain observation bin can be calculated as 5 ÷ M. Considering that if the sample proportion of a certain observation bin is too small, the sample size will also be too small, which will affect the statistical reliability of subsequent evidence weight values ​​and information values. Therefore, it is necessary to ensure that each observation bin in the final binning result corresponds to a sufficient amount of sample data through the merging process. In addition, the third preset threshold can be preset to 2%, 5%, or any pure decimal between 2% and 5%.

[0043] In step S44, in order to maintain or shape the monotonic or ordered relationship between the binned variables and the target variable (e.g., default rate) to the greatest extent possible during the merging process (this is the lifeline to ensure the logical correctness and interpretability of the subsequent model, and to avoid the forced merging of high-risk customer groups with low-risk customer groups, which would lead to a decrease in the discriminative ability of the subsequent model), it is further preferred that, based on the comparison result of the sample proportion and the third preset threshold, the binning of multiple observations of any variable is merged to obtain the final binning result of any variable and the sample proportion of each bin is greater than or equal to the third preset threshold, including but not limited to the following steps S441 to S442.

[0044] S441. Traverse each observation bin of any variable as follows: Calculate the sample percentage of the currently traversed bin. If the sample percentage is less than a third preset threshold, first calculate the first sample default rate of the currently traversed bin and the second sample default rate of at least one observation bin adjacent to the currently traversed bin. Then merge the currently traversed bin with an observation bin in the at least one observation bin that has the second sample default rate closest to the first sample default rate to obtain a new observation bin for any variable. Then traverse the next observation bin. Otherwise, directly traverse the next observation bin. Finally, after traversal is completed, execute step S442. Here, the sample percentage refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M. The first sample default rate and the second sample default rate refer to the ratios of the number of defaulting samples to the total number of samples corresponding to the corresponding observation bin.

[0045] In step S441, continuing with the example in step S43, if the currently traversed bin is a certain observation bin, such that it corresponds to a total of 5 sample data (i.e., the sample size is equal to 5), and among these 5 sample data, one sample data has a dependent variable value of "1", then the first sample default rate of the currently traversed bin can be calculated as 1 ÷ 5 = 20%; similarly, the default rates of the two observation bins adjacent to the currently traversed bin—the first observation bin [18, 25] and the second observation bin (35, 45]—can be calculated. If the second sample default rate of the first observation bin [18,25] is 21%, and the second sample default rate of the second observation bin (35,45) is 40%, then the first observation bin [18,25] and the currently traversed bin (25,35] are merged to obtain a new observation bin [18,35]. Furthermore, if the currently traversed bin is the first observation bin [18,25], then only one observation bin is adjacent to the currently traversed bin—the observation bin (25,35).

[0046] S442. Determine whether the sample proportion of each observation bin of any variable is greater than or equal to the third preset threshold. If so, take each observation bin of any variable as the final binning result; otherwise, return to step S441.

[0047] In addition, for each of the N cleaned independent variables that are discrete, each unique value can be binned as an observation. Then, based on the aforementioned steps S441 to S442, the binning of multiple observations of the corresponding variable is merged to obtain the final binning result where the sample proportion of each bin is greater than or equal to the third preset threshold. This also achieves the goal of maximizing the maintenance or shaping of the monotonic or ordered relationship between the binned variable and the target variable (e.g., default rate).

[0048] S5. For each variable, based on the binning results or discreteness of the corresponding variable and the dependent variable in the M sample data, calculate the information value of the corresponding variable, wherein the information value is used to positively measure the predictive ability of the corresponding variable for loan default.

[0049] In step S5, the larger the information value, the stronger the predictive ability of the variable, and the easier it is to be selected by the model, thus it can be used for variable screening later. Specifically, for each variable, the information value of the corresponding variable is calculated based on the binning result or discreteness of the corresponding variable and the dependent variable in the M sample data, including but not limited to the following steps S51 to S53.

[0050] S51. For a certain variable among the N cleaned independent variables, if the corresponding variable was previously determined to be a continuous variable, then obtain the corresponding binning result.

[0051] S52. For each observation in the binning result of the certain variable, binning is performed, and the corresponding evidence weight value is calculated according to the following formula based on all the corresponding sample data and the dependent variable in the M sample data:

[0052] In the formula, Represents positive integers. This indicates the first binning result of a certain variable. The evidence weight value for each bin of observations This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of default samples in all sample data corresponding to each observation bin. This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of non-default samples in all sample data corresponding to each observation bin. This represents the total number of default samples determined based on the dependent variable from the M sample data. This represents the total number of non-default samples determined based on the dependent variable in the M sample data.

[0053] In step S52, referring again to the example in step S441, if the first A bin for each observation is defined such that it corresponds to a total of 5 sample data points (i.e., the sample size is equal to 5), and among these 5 sample data points, one sample data point has a dependent variable value of "1". Equals 1, It equals 5 - 1 = 4. Furthermore, and The sum of these is the total number of samples M.

[0054] S53. Based on the evidence weight values ​​of each bin for the observed values, the information value of a certain variable is calculated according to the following formula. :

[0055] In the formula, This represents the total number of bins in the binning results of a certain variable, and the information value is used to positively measure the predictive power of the corresponding variable for post-loan default.

[0056] In step S53, as seen from the above formula, the information value primarily measures the predictive ability through the difference between the default rate and the non-default rate. Furthermore, for each of the N cleaned independent variables that is discrete, each unique value can be binned as an observation based on the discreteness of the corresponding variable, and then the corresponding information value can be calculated based on steps S52-S53 above. (The binning results may need to be merged and processed based on the aforementioned steps S441 to S442 before calculation).

[0057] S6. For each pair of variables among all the cleaned independent variables, calculate the correlation coefficient between the two variables based on the binning or discreteness of the two variables, and remove the variable with the smaller information value in the two variables when the absolute value of the correlation coefficient is greater than or equal to the first preset threshold.

[0058] In step S6, the correlation coefficient is used to measure the degree of correlation between two variables, and can be calculated, but is not limited to, according to the following formula:

[0059] In the formula, Indicates the first variable With the second variable The correlation coefficient between them Indicates based on the first variable The binning result or discrete case and the second variable The covariance obtained from the binning process or the conventional calculation of the discrete case. Indicates based on the first variable The variance obtained from conventional calculations of binning results or discrete cases. Indicates based on the second variable The variance is calculated conventionally based on the binning results or discrete cases. Considering that an absolute value of the correlation coefficient greater than or equal to 0.8 generally indicates a strong correlation between two variables, it is necessary to retain variables with larger information values ​​and remove variables with smaller information values ​​to facilitate subsequent model training. Specifically, when the absolute value of the correlation coefficient is greater than or equal to a first preset threshold (e.g., 0.8), variables with smaller information values ​​among the corresponding two variables are removed. Furthermore, considering that an information value less than 0.02 generally indicates a weak predictive ability of the variable for the target variable, it is necessary to remove such variables. Therefore, to reduce the computational resource overhead of step S6, preferably, after calculating the information values ​​of each variable and before calculating the correlation coefficient between the two variables, the method further includes: removing variables with information values ​​less than a fourth preset threshold (e.g., 0.02) from the N cleaned independent variables.

[0060] S7. Import all the remaining cleaned independent variables and dependent variables from the M sample data into the machine learning model for model training to obtain the loan default prediction model and output it.

[0061] In step S7, all remaining cleaned independent variables serve as model inputs, and the dependent variable serves as model outputs or sample labels. The loan default prediction model can be trained using a conventional calibration and verification modeling process. Furthermore, the machine learning model is an existing model, and can be constructed using, but is not limited to, machine learning algorithms suitable for binary classification problems, such as logistic regression, gradient boosting decision trees (e.g., XGBoost or LightGBm), or random forests.

[0062] Therefore, based on the end-to-end automated pipeline credit modeling method described in steps S1 to S7 above, a new solution is provided that can achieve end-to-end and fully automated processing from raw data to credit model in the specific scenario of inclusive finance. First, a large amount of sample data is acquired and automatically cleaned and variable type determined. Then, for continuous variables, a decision tree algorithm is used for automated binning. Next, based on the binning results or discrete cases, the information value of each variable and the correlation coefficient between variables are automatically calculated, and variables are automatically filtered according to preset thresholds. Finally, the remaining variables obtained from the filtering and the original dependent variable are imported into a machine learning model for model training, resulting in a loan default prediction model, which is then output. This achieves full automation from raw data to model deployment, significantly improving the modeling efficiency, objectivity, and reproducibility in multi-dimensional and big data scenarios of inclusive finance, facilitating practical application and promotion.

[0063] like Figure 2 As shown, the second aspect of this embodiment provides a virtual system for implementing the end-to-end automated pipeline credit modeling method described in the first aspect, including a sample data acquisition unit, a data cleaning and processing unit, a variable type judgment unit, a decision binning processing unit, an information value calculation unit, a variable correlation screening unit, and a prediction model training unit; The sample data acquisition unit is used to acquire M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data contains N independent variables used as model inputs and one dependent variable used as model outputs, where N represents a positive integer greater than 100. The N independent variables are used to reflect the multi-dimensional information of the lender before the loan is issued, and the dependent variable is used to indicate whether the lender has defaulted after the loan is issued. The data cleaning and processing unit is communicatively connected to the sample data acquisition unit. It is used to perform data cleaning processing on the N independent variables in the corresponding data for each of the M sample data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes text-to-numerical processing and missing value processing. The variable type determination unit is communicatively connected to the data cleaning and processing unit. It is used to determine whether a variable is a discrete variable or a continuous variable for each variable among the N cleaned independent variables based on all the observations of the corresponding variable. The observations include M observations that correspond one-to-one with the M sample data. The decision binning processing unit is communicatively connected to the variable type determination unit. It is used to automatically bin the corresponding variable using a decision tree algorithm if any variable among the N cleaned independent variables is determined to be a continuous variable, and to obtain the binning processing result of the corresponding variable. The information value calculation unit is communicatively connected to the data cleaning and processing unit and the decision binning and processing unit, respectively. It is used to calculate the information value of each variable based on the binning and processing result or discreteness of the corresponding variable and the dependent variable in the M sample data. The information value is used to positively measure the predictive ability of the corresponding variable for loan default. The variable correlation screening unit is communicatively connected to the data cleaning and processing unit, the decision binning and processing unit, and the information value calculation unit. It is used to calculate the correlation coefficient between the two variables for each pair of variables in all the cleaned independent variables, based on the binning and processing results or the discreteness of the two variables. When the absolute value of the correlation coefficient is greater than or equal to a first preset threshold, the variable with a smaller information value in the two variables is removed. The prediction model training unit is communicatively connected to the variable correlation screening unit, and is used to import all the remaining cleaned independent variables and dependent variables from the M sample data into the machine learning model for model training, so as to obtain the loan default prediction model and output it.

[0064] The working process, working details and technical effects of the aforementioned system provided in the second aspect of this embodiment can be found in the end-to-end automated pipeline credit modeling method described in the first aspect, and will not be repeated here.

[0065] like Figure 3 As shown, the third aspect of this embodiment provides a computer device for executing the end-to-end automated pipeline credit modeling method as described in the first aspect. The device includes a storage module, a processing module, and a transceiver module connected in sequence. The storage module stores a computer program, the transceiver module sends and receives messages, and the processing module reads the computer program and executes the end-to-end automated pipeline credit modeling method as described in the first aspect. Specifically, the storage module may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; the processing module may, but is not limited to, use a microprocessor of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power supply module, a display screen, and other necessary components.

[0066] The working process, working details and technical effects of the aforementioned computer equipment provided in the third aspect of this embodiment can be found in the end-to-end automated pipeline credit modeling method described in the first aspect, and will not be repeated here.

[0067] This fourth aspect of the embodiment provides a computer-readable storage medium storing instructions comprising the end-to-end automated pipeline credit modeling method as described in the first aspect. Specifically, the computer-readable storage medium stores instructions that, when executed on a computer, perform the end-to-end automated pipeline credit modeling method as described in the first aspect. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0068] The working process, working details and technical effects of the aforementioned computer-readable storage medium provided in the fourth aspect of this embodiment can be found in the end-to-end automated pipeline credit modeling method described in the first aspect, and will not be repeated here.

[0069] This fifth aspect of the embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, implement the end-to-end automated pipeline credit modeling method as described in the first aspect. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0070] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An end-to-end automated pipeline credit modeling method for inclusive finance scenarios, characterized in that, include: Obtain M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data contains N independent variables as model inputs and one dependent variable as model outputs, where N represents a positive integer greater than 100. The N independent variables reflect multi-dimensional information about the lender before the loan is issued, and the dependent variable indicates whether the lender defaults after the loan is issued. For each sample data in the M sample data, data cleaning processing is performed on the N independent variables in the corresponding data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes text-to-numerical processing and missing value processing. For each variable among the N cleaned independent variables, determine whether the corresponding variable is a discrete variable or a continuous variable based on all the observed values ​​of the corresponding variable, wherein the all observed values ​​include M observed values ​​that correspond one-to-one with the M sample data. For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, the decision tree algorithm is used to automatically bin the corresponding variable to obtain the binning result of the corresponding variable. For each variable, based on the binning results or discreteness of the corresponding variable and the dependent variable in the M sample data, the information value of the corresponding variable is calculated, wherein the information value is used to positively measure the predictive ability of the corresponding variable for loan default. For each pair of variables among all the cleaned independent variables, the correlation coefficient between the two variables is calculated based on the binning or discreteness of the two variables. When the absolute value of the correlation coefficient is greater than or equal to a first preset threshold, the variable with a smaller information value among the two variables is removed. The remaining cleaned independent variables and dependent variables from the M sample data are imported into a machine learning model for model training to obtain a loan default prediction model, which is then output.

2. The end-to-end automated pipeline credit modeling method according to claim 1, characterized in that, For each of the N cleaned independent variables, based on all observed values ​​of the corresponding variable, determine whether the corresponding variable is a discrete or continuous variable, including: For each variable among the N cleaned independent variables, based on all the observed values ​​of the corresponding variable, the number of unique values ​​of the corresponding variable is counted. Then, it is determined whether the number of unique values ​​is less than or equal to a second preset threshold. If so, the corresponding variable is determined to be a discrete variable; otherwise, the corresponding variable is determined to be a continuous variable. Here, all the observed values ​​include M observed values ​​that correspond one-to-one with the M sample data. The number of unique values ​​refers to the total number of non-repeating values ​​obtained after deduplication of all the observed values ​​of the corresponding variable.

3. The end-to-end automated pipeline credit modeling method according to claim 1, characterized in that, For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, a decision tree algorithm is used to automatically bin the corresponding variable, obtaining the binning result of the corresponding variable, including: For any variable among the N cleaned independent variables, if the corresponding variable is determined to be a continuous variable, then all observations of the corresponding variable are obtained, wherein the all observations include M observations that correspond one-to-one with the M sample data. By applying all the observed values ​​of any variable and the dependent variable in the M sample data, and with the AUC value of the decision tree model as the optimization objective, the multiple hyperparameters of the decision tree algorithm are optimized to obtain the optimal values ​​of the multiple hyperparameters, wherein the multiple hyperparameters include the minimum number of sample leaf nodes, the complexity pruning parameter and / or the maximum branch depth. The optimal values ​​of the multiple hyperparameters are substituted into the decision tree algorithm to obtain the optimal decision tree algorithm. The optimal decision tree algorithm is then applied to bin the observations of any variable to obtain multiple bins of the observations of any variable. Based on the comparison results of the sample proportion and the third preset threshold, the bins of multiple observations of any variable are merged to obtain the final binning result of any variable where the sample proportion of each bin is greater than or equal to the third preset threshold. The sample proportion refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M.

4. The end-to-end automated pipeline credit modeling method according to claim 3, characterized in that, Based on the comparison result between the sample proportion and the third preset threshold, the multiple observations of any variable are binned and merged to obtain the final binning result where the sample proportion of each bin is greater than or equal to the third preset threshold. This includes the following steps S441 to S442: S441. Traverse each observation bin of any variable as follows: Calculate the sample percentage of the currently traversed bin. If the sample percentage is less than a third preset threshold, first calculate the first sample default rate of the currently traversed bin and the second sample default rate of at least one observation bin adjacent to the currently traversed bin. Then merge the currently traversed bin with an observation bin in the at least one observation bin that has the second sample default rate closest to the first sample default rate to obtain a new observation bin for any variable. Then traverse the next observation bin. Otherwise, directly traverse the next observation bin. Finally, after traversal is completed, execute step S442. Here, the sample percentage refers to the ratio of the sample size corresponding to the corresponding observation bin to the total number of samples M. The first sample default rate and the second sample default rate refer to the ratio of the number of defaulting samples to the sample size corresponding to the corresponding observation bin, respectively. S442. Determine whether the sample proportion of each observation bin of any variable is greater than or equal to the third preset threshold. If so, take each observation bin of any variable as the final binning result; otherwise, return to step S441.

5. The end-to-end automated pipeline credit modeling method according to claim 1, characterized in that, For each variable, based on the binning results or discreteness of the corresponding variable and the dependent variable in the M sample data, the information value of the corresponding variable is calculated, including: For any variable among the N cleaned independent variables, if the corresponding variable was previously determined to be a continuous variable, then obtain the corresponding binning result. For each observation in the binning results of a certain variable, the corresponding evidence weight value is calculated according to the following formula based on all corresponding sample data and the dependent variable in the M sample data: In the formula, Represents positive integers. This indicates the first binning result of a certain variable. The evidence weight value for each bin of observations This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of default samples in all sample data corresponding to each observation bin. This indicates that the dependent variable, determined based on the M sample data, is related to the first... The number of non-default samples in all sample data corresponding to each observation bin. This represents the total number of default samples determined based on the dependent variable from the M sample data. This represents the total number of non-default samples determined based on the dependent variable in the M sample data; Based on the evidence weight values ​​of each bin for the observed values, the information value of a certain variable is calculated according to the following formula. : In the formula, This represents the total number of bins in the binning results of a certain variable, and the information value is used to positively measure the predictive power of the corresponding variable for post-loan default.

6. The end-to-end automated pipeline credit modeling method according to claim 1, characterized in that, After calculating the information values ​​of each variable and before calculating the correlation coefficient between the two variables, the method further includes: Remove variables whose information values ​​are less than a fourth preset threshold from the N cleaned independent variables.

7. An end-to-end automated pipeline credit modeling system for inclusive finance scenarios, characterized in that, It includes a sample data acquisition unit, a data cleaning and processing unit, a variable type judgment unit, a decision binning processing unit, an information value calculation unit, a variable correlation screening unit, and a prediction model training unit; The sample data acquisition unit is used to acquire M sample data for the inclusive finance scenario, where M represents a positive integer greater than 1000. The sample data contains N independent variables used as model inputs and one dependent variable used as model outputs, where N represents a positive integer greater than 100. The N independent variables are used to reflect the multi-dimensional information of the lender before the loan is issued, and the dependent variable is used to indicate whether the lender has defaulted after the loan is issued. The data cleaning and processing unit is communicatively connected to the sample data acquisition unit. It is used to perform data cleaning processing on the N independent variables in the corresponding data for each of the M sample data to obtain the corresponding N cleaned independent variables. The data cleaning processing includes text-to-numerical processing and missing value processing. The variable type determination unit is communicatively connected to the data cleaning and processing unit. It is used to determine whether a variable is a discrete variable or a continuous variable for each variable among the N cleaned independent variables based on all the observations of the corresponding variable. The observations include M observations that correspond one-to-one with the M sample data. The decision binning processing unit is communicatively connected to the variable type determination unit. It is used to automatically bin the corresponding variable using a decision tree algorithm if any variable among the N cleaned independent variables is determined to be a continuous variable, and to obtain the binning processing result of the corresponding variable. The information value calculation unit is communicatively connected to the data cleaning and processing unit and the decision binning and processing unit, respectively. It is used to calculate the information value of each variable based on the binning and processing result or discreteness of the corresponding variable and the dependent variable in the M sample data. The information value is used to positively measure the predictive ability of the corresponding variable for loan default. The variable correlation screening unit is communicatively connected to the data cleaning and processing unit, the decision binning and processing unit, and the information value calculation unit. It is used to calculate the correlation coefficient between the two variables for each pair of variables in all the cleaned independent variables, based on the binning and processing results or the discreteness of the two variables. When the absolute value of the correlation coefficient is greater than or equal to a first preset threshold, the variable with a smaller information value in the two variables is removed. The prediction model training unit is communicatively connected to the variable correlation screening unit, and is used to import all the remaining cleaned independent variables and dependent variables from the M sample data into the machine learning model for model training, so as to obtain the loan default prediction model and output it.

8. A computer device, characterized in that, It includes a storage module, a processing module, and a transceiver module that are sequentially connected in communication. The storage module is used to store computer programs, the transceiver module is used to send and receive messages, and the processing module is used to read the computer programs and execute the end-to-end automated pipeline credit modeling method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that... The computer-readable storage medium stores instructions that, when executed on a computer, perform the end-to-end automated pipeline credit modeling method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or the instructions are executed by the computer, they implement the end-to-end automated pipeline credit modeling method as described in any one of claims 1 to 6.

Images