TS-WRSVM: twin structural weighted relaxed support vector machine. Issue 3 (3rd July 2019)

• Record Type: Journal Article
• Authors: Sheykh Mohammadi, Fatemeh; Amiri, Ali
• Is Part Of: Connection science. Volume 31:Issue 3(2019)
• Page Start: 215
• Page End: 243
• Publication Date: 2019-07-03
• Subjects: Imbalanced data classification -- cost-sensitive learning -- structured learning -- TS-WRSVM -- outliers; Neural computers -- Periodicals; Artificial intelligence -- Periodicals; Cognitive science -- Periodicals; Connectionism -- Periodicals; 006.3
• DOI: 10.1080/09540091.2019.1573418
• Languages: English
• ISSNs: 0954-0091

Abstract

ABSTRACT: Classification of data with imbalanced class distributions is a major problem in the data mining community. Imbalanced classification is a challenging task in the presence of outliers. In this paper, we propose a new cost-sensitive learning method with regard to the structure of data distribution for classifying imbalanced data and diminishing the effect of outliers. The proposed method combines the benefits of "structured" learning models (such as structural support vector machine) with the advantages of "cost-sensitive" learning models (such as weighted relaxed support vector machine). We call our method twin structural weighted relaxed support vector machine (TS-WRSVM) . A TS-WRSVM uses two nonparallel hyperplanes to determine the class label of new data so that each model only considers the structural information of one class. We allocate a weight and a limited amount of penalty-free slack to each model by considering the size of each class. Results of experiments indicate that a TS-WRSVM is superior to other current algorithms based on cost-sensitive learning in the areas of classification accuracy and computational time.