Mutual information-based data discretization and feature selection integrated method and apparatus

Abstract

The invention discloses a mutual information-based data discretization and feature selection integrated method. The method comprises the steps of 1) generating a proper candidate breakpoint set by performing breakpoint analysis on data; 2) searching for an optimal breakpoint subset in the candidate breakpoint set by adopting forward search, and evaluating the breakpoint subset by calculating mutual information between a data division result and original tag distribution of the data; 3) defining a search stop condition that the information gain ratio is less than a preset threshold or the total number of selected breakpoints exceeds the preset threshold; and 4) performing discretization and feature selection on the data by using the optimal breakpoint subset. The invention furthermore discloses a mutual information-based data discretization and feature selection integrated apparatus. According to the mutual information-based data discretization and feature selection integrated method and apparatus, the data discretization and feature selection processes are organically integrated, so that unrelated and redundant information in the data can be effectively removed and the performance of a subsequent learning algorithm can be improved.

Description

technical field [0001] The invention belongs to the field of data preprocessing in data mining, adopts mutual information as an evaluation criterion, and integrates data discretization and feature selection. The present invention can be directly applied to the preprocessing stage of continuous data with label information, and simultaneously completes discretization and feature selection of data. The invention relates to a data discretization and feature selection integration method and device based on mutual information. Background technique [0002] Data discretization is a process of converting continuous features into nominal data or ordered data. Divide the value range of continuous features into multiple small intervals, and each interval represents an ordered numerical or categorical data. These discretized data are suitable for some learning algorithms that cannot use continuous features. At the same time, they can effectively remove hidden defects in the data and i...

AI Technical Summary

Problems solved by technology

[0004] In the process of data preprocessing, the existing technology regards the discretization of continuous data and feature selection as two independent sub-processes, ignoring the internal relationship between the two, resulting in more information loss and lower follow-up Performance of Learning Algorithms

Images