Non-unique Decision Differentiation Entropy-based Feature Selection Approach

Feature selection has become one of hot topics in the field of machine learning research.The core of the feature selection is to obtain a subset of features from the original features that can retain the initial physical features according to a certain evaluation criterion.Rough set theory is a mathematical tool to characterize the degree of inconsistency of categorical data and has been successfully applied to the processing and analysis of symbolic data.However,there is no systematic research on the problem of complex data mining that coexist with widely existing symbols,values and fuzzy variables.However,the fuzzy rough set feature selection method can effectively deal with the symbol,numerical data and the mixed data of the two,so that the improvement can make the researchers to obtain the learning model which is simple and easy to understand.In this paper,the research on rough set theory needs to divide the data set equally,find out the positive or boundary domain in the set,and use the dependence or uncertainty of attributes to construct the feature selection algorithm.This calculation of the use of equivalent partitioning to solve uncertainty is complicated and not conducive to use in larger data sets.Therefore,based on the rough set of symbolic data,this paper proposes a feature selection algorithm based on non-unique decision differentiation entropy.Firstly,three different non-unique decision-making methods are used to obtain the information of the boundary of the attribute set,and then the difference entropy is used to evaluate the importance of the attribute subset to obtain the reduction result.The calculation of the non-unique decision simplifies the rough set inconsistency of the process of solving.Secondly,this paper deals with the non-unique decision-making model based on fuzzy similarity relation in the coexistence of symbolic and numerical data.A feature selection algorithm based on non-unique decision differentiation entropy of mixed data is obtained.The accuracy of the premise simplifies the calculation of uncertainty in hybrid information systems.The experimental results in this paper are based on eight kinds of symbol data sets and nine kinds of mixed data sets for feature selection.The feature selection algorithm of non-unique decision differentiation entropy is compared with other algorithms in terms of classification accuracy and AUC.The algorithm includes ACO,FRFS,PCA and PSO.Experimental results show that the feature subset obtained by feature selection algorithm based on non-unique decision differentiation entropy has high classification accuracy,which proves that the proposed algorithm has practical significance.