The Research of One Novel Cost‐Sensitive Classification Algorithm. Issue 5 (1st September 2018)

• Record Type: Journal Article
• Authors: Zhou, Jingjing; Sun, Weifeng; Han, Xiaomin; Lu, Ruqiang; Zhang, Yuanqi; Zhang, Shenwei
• Is Part Of: Chinese journal of electronics. Volume 27:Issue 5(2018)
• Page Start: 1015
• Page End: 1024
• Publication Date: 2018-09-01
• Subjects: Graph‐based semi‐supervised classification (GSSC) -- Cost‐sensitive -- Rescale; data mining -- graph theory -- learning (artificial intelligence) -- pattern classification; synthetic minority oversampling technique algorithm -- misclassification costs -- traditional Graph -- semisupervised classification algorithms -- high classification accuracy -- finance -- medicine -- global classification accuracy -- global misclassification -- cost‐sensitive classification algorithm -- local consistency -- global consistency -- classification accuracy -- cost‐sensitive classification; Electronics -- Periodicals; Electronics -- China -- Periodicals; Electronics; China; Periodicals; 621.38105
• DOI: 10.1049/cje.2018.01.002
• Languages: English
• ISSNs: 1022-4653

Abstract

Abstract : Assuming that misclassification costs between diffierent categories are equal, traditional Graph based semi‐supervised classification (GSSC) algorithms pursues high classification accuracy. In many practical problems, especially in the fields of finance and medicine, compared with global classification accuracy, less cost on global misclassification is more likely to be the most significant factor. We propose one novel cost‐sensitive classification algorithm based on the local and global consistency, which utilizes the semi‐supervised classification algorithms better, and ensures higher classification accuracy on the basis of reducing overall cost. Our improved algorithm may bring some problems due to unbalanced data account, so we introduce synthetic minority oversampling technique algorithm for further optimization. Experimental results of bank loans and medical problems verify the effiectiveness of our novel classification algorithm.