Enhanced Category-Feature Association Measure

Text classification is one of the severe challenges for categorizing large and high-dimensional text data accurately and efficiently. Many features confuse the classification process, and feature selection (FS) strategies should be used to deal with the problem of high dimensionality. This paper proposes a novel FS technique based on enhanced category-feature association measure (ECFAM). ECFAM utilizes the existence and elimination of terms and the complicated relationships among the terms across different sections. This one-of-a-kind approach emphasizes the key role of ancillary terms in classifying and differentiating categories. The comparison is done on two important datasets, Reuters-21578 and 20-Newsgroups, through two widely employed supervised machine learning classifiers and one deep learning algorithm. Throughout our experiments, we investigate the feature sizes in nine different feature sets, ranging from 50 to 4000. Experimental data show that ECFAM always performs better than other methods concerning accuracy and computational cost.