Score card model optimization method based on decision tree feature fusion

Abstract

The invention discloses a score card model optimization method based on decision tree feature fusion, and the method comprises the steps: S1, screening feature information based on filling informationand credit investigation information of a user; taking screened feature information as a feature variable; S2, replacing the variable value of the characteristic variable in the step S1 with badrate,and converting a category type variable in the characteristic variable into a numerical value type variable; S3, converting the data into numerical variables in the step S2, generating a tree structure by using a decision tree method, performing variable fusion on the numerical variables, and extracting values of all segmentation points of each numerical variable in the tree structure so as to ensure the reliability of the information; S4, representing the numerical result of each branch of the decision tree in the step S3 into a new information variable by adopting a way of ware coding; andS5, synthesizing the new information variable after the way coding in the step S4 with other original information variables to serve as a logistic regression input value to establish a score card model. According to the invention, interpretability in business meaning and optimal fusion in data meaning are both considered.

Description

technical field [0001] The invention relates to the technical field of the Internet, in particular to a scoring card model optimization method based on decision tree feature fusion. Background technique [0002] Currently, a credit scoring model is a model that evaluates the credit status of a loan applicant to predict the applicant's probability of serious default or bad debt in the future. [0003] The scorecard model is the most important and commonly used credit scoring model, and the generation of feature variables has the greatest impact on the effect of the model. For the scorecard model, the general feature generation method comes from the information filled in by users, credit information, tripartite Information or other information sources, through the processing of fields in these information, many characteristic variables are derived, but these characteristic variables are relatively independent in the scorecard model, and there are certain correlations between s...

AI Technical Summary

Problems solved by technology

[0005] (1) The single feature derivation method traditionally used only considers the information of a single feature, but does not take into account the relationship between features, which is neglected in the actual business sense and does not really achieve better derivation in the sense of data

[0006] (2) The traditional method does not select the optimal segmentation point according to the data performance in the selection of the variable segmentation point, and there will be a large loss in the information value IV (information value) of the final segmentation result, which will affect the final effect of the model

[0007] (3) In addition, the traditional method is relatively subjective and does not have uniform applicability, and the effects made by different modeling engineers are unstable

[0008] (4) The traditional method takes a long time to work, and must rely on the manual subjective operation of modeling experts, which cannot realize the automatic establishment of machine learning models

Images