Model training system based on separation degree index

Abstract

The invention relates to a model training system based on a separation degree index. The model training system comprises a model training unit, a model pruning and compressing unit and an output unit.The model training unit comprises the following modules of: a, data cleaning module for original variable cleaning; b, a feature selection module for screening candidate feature sets compressed by amodel; c, a model training module for model training and optimization. The model pruning and compressing unit comprises the following modules of: d, a data sample grouping module for data sample grouping; e, a feature correlation discrimination module used for calculating correlation coefficients of features and target variables and grouping and sorting samples; f, a feature optimal breakpoint selection module for selecting the optimal breakpoints of the features; g, a feature separation degree index calculation module which constructs feature separation degree indexes and outputs a feature with the best effect. The output unit comprises the following modules of: h, an optimal feature selection module for optimal feature selection; and i, an output module used for outputting a single-pointrule list. According to the method, the established model can be trained under the condition that the data of one party is not transmitted out, so that the data security and customer privacy of two parties are effectively protected.

Description

technical field [0001] The invention relates to a training system of a machine learning model, specifically a model training system based on a separation index. Background technique [0002] In the field of machine learning and artificial intelligence, the traditional joint modeling method is generally that both partners hold part of the data (explanatory variables or labels), and one party carries the data to the other party for data cleaning, processing, and modeling deployment. There are two problems in such a process. One is that the direct carrying of data may lead to the risk of data leakage, and the other is the risk of legal compliance. With the increasingly stringent privacy protection legislation of citizens, the circulation of sensitive data may touch legal issues. lead to regulatory intervention. [0003] For the above problems, the existing solution is called federated learning. The core idea of ​​federated learning is data encryption. The two parties independe...

AI Technical Summary

Problems solved by technology

In addition, model training with encrypted data requires redevelopment of the model training code. The efficiency of model training is much lower than that of local plaintext training. As for the asynchronous optimization of parameters during model training, it will further lead to loss of model effect.

Images