A multi-objective genetic algorithm for text feature selection using the relative discriminative criterion. (1st July 2020)

• Record Type: Journal Article
• Authors: Labani, Mahdieh; Moradi, Parham; Jalili, Mahdi
• Is Part Of: Expert systems with applications. Volume 149(2020)
• Publication Date: 2020-07-01
• Subjects: Text classification -- Feature selection -- Multi-objective optimization -- Relative Discriminative Criterion -- Relevancy -- Redundancy; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2020.113276
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: A multi-objective feature selection method called MORDC is proposed for text classification tasks. The relevancy of features is computed by using the RDC metric, which is specifically designed for text categorization tasks. The relevancy of features is computed by using the text specific RDC metric MORDC selects maximum relevant and minimum redundant features. MORDC does not employ any learning model to evaluate the effectiveness of selected features. The results show that MORDC outperforms the other methods. Abstract: With exponentially increasing the number of digital documents, text classification has become a major task in data science applications. Selecting discriminative features highly relevant to class labels while having low levels of redundancy is essential to improve the performance of text classification methods. In this paper, we propose a novel multi-objective algorithm for text feature selection, called Multi-Objective Relative Discriminative Criterion (MORDC), which balances minimal redundant features against those maximally relevant to the target class. The proposed method employs a multi-objective evolutionary framework to search through the solution space. The first objective function measures the relevance of the text features to the target class, whereas the second one evaluates the correlation between the features. None of these objectives use learning to evaluate the goodness of the selected features; thus, the proposed method can be classified as a multivariate filter method. In order to assess the effectiveness of the proposed method, several experiments are performed on three real-world datasets. Comparisons with state-of-the-art feature selection methods show that in most cases MORDC results in better classification performance.