FE-TAC: an effective document classification method combining feature extraction and feature selection

Abstract

An effective classification method requires the most informative and relevant set of features. In this paper, we discuss an enhanced text classification method combining feature extraction (FE) and feature selection. First, we used the FE method to extract features from text data and then apply the feature selection method to select the most relevant features out of those extracted features. During feature selection, we introduce a new measure called term affinity to the class (TAC) to estimate the degree of retaining capability of the term as a member of the particular class. TAC is computed based on the combination of normalise document frequency and summing up the occurrence frequency of the term to the specific class. Experimental results on three existing datasets - BBC, Classic4, 20 Newsgroup, and our own dataset called 'Sangai' show that the proposed method outperforms the other competent methods in terms of accuracy.