Model calibration method and device

Abstract

One or more embodiments of the invention disclose a model calibration method and a model calibration device, which are used for realizing automatic model calibration and improving the precision of model calibration. The method comprises the steps of obtaining an updating model corresponding to an initial model; obtaining a verification set for calibrating the updating model, wherein the verification set comprises a plurality of pieces of sample data; inputting the sample data into the updating model to obtain an output set corresponding to the verification set, wherein the output set comprisespredicted values respectively corresponding to the sample data; searching X segmentation points from each predicted value according to a specified segmentation point search algorithm; and dividing the output set into a plurality of segmentation intervals by using the segmentation points, wherein the value of X is smaller than the number of the predicted values; and determining a calibration result corresponding to each subsection interval according to the predicted value in each subsection interval and the index distribution of the initial model in each subsection interval.

Description

technical field [0001] This description relates to the technical field of data processing, in particular to a model calibration method and device. Background technique [0002] With the development of big data analysis and machine learning theory, more and more enterprises use machine learning technology to build algorithm models in business development for analysis and prediction in business scenarios. Especially in the risk control scenario, various machine learning, deep learning and other algorithms are gradually popularized, which greatly improves the risk control effect in various businesses. [0003] The output score of the risk control model usually needs to reflect the real business risk concentration or risk probability. After the general model is launched, it will set a specific model score threshold based on historical data analysis and business requirements, and carry out risk interception or release. For example, the access scorecard model used in the credit ...

AI Technical Summary

Problems solved by technology

[0004] However, the risk situation faced in various business scenarios usually has a certain timeliness, and due to the confrontation of risk attack and defense, the risk situation is constantly changing and migrating

This characteristic of risk business also determines that the risk control model needs to be updated iteratively frequently, and frequent model updates will bring greater workload to risk operators and easily lead to operational risks of strategy adjustments

Images