Distributed score card model establishing method

Abstract

The distributed score card model establishing method comprises the following steps: binning sample data of each label variable into N sample intervals, and calculating chi-square values of adjacent intervals; combining the adjacent intervals corresponding to the minimum chi-square value into a new interval, judging whether the number of the combined intervals reaches the target number or not, if so, executing S3, or otherwise, executing S4; S3, carrying out monotonicity testing on the chi-square values, calculated by the combined intervals, of the adjacent intervals, if the chi-square values are monotonous, executing the step S5, and if not, combining the adjacent intervals corresponding to the minimum chi-square value into a new interval, and executing the step S3; S4, recalculating the chi-square value of each combined adjacent interval and executing the step S2; and S5, judging whether the proportion of the sample data of the interval exceeds a set proportion value, if so, abandoning the label variable, otherwise, calculating an information credit investigation value of the interval, selecting the label variable of which the information credit investigation value reaches a preset requirement as a target label variable, and substituting the sample data of the target label variable into model training.

Description

technical field [0001] The invention relates to the technical field of scorecard model building, in particular to a distributed scorecard model building method. Background technique [0002] A credit scorecard is a tool that uses the performance of previous applicants to predict the behavior of new applicants. Credit scorecard technology uses historical data and statistical techniques to analyze the degree to which various risk factors affect default rates, resulting in a score by which banks can classify the riskiness of loan applications. According to the construction method, credit scorecards can be divided into three types: judgment scorecards (expert scorecards), statistical scorecards and hybrid scorecards. [0003] With the explosive development of Internet finance, credit score card technology is widely used in the risk control systems of various Internet finance companies. At this stage, credit score cards are divided into application score cards and behavior scor...

AI Technical Summary

Problems solved by technology

[0005] Insufficient model training: Due to the limitation of running memory, the scorecard model is used for training data samples are limited, resulting in a suboptimal model

[0006] The model takes a long time: During the model training of the stand-alone version of the scoring card, due to the slow running speed of the stand-alone machine, the running time is very long, and it also takes a long time to calculate the score when outputting the score, which cannot respond quickly and cannot meet the business needs. need

Images