Regularized max-min linear discriminant analysis. (June 2017)

• Record Type: Journal Article
• Authors: Shao, Guowan; Sang, Nong
• Is Part Of: Pattern recognition. Volume 66(2017:Jun.)
• Page Start: 353
• Page End: 363
• Publication Date: 2017-06
• Subjects: MMLDA max-min linear discriminant analysis -- RMMLDA regularized max-min linear discriminant analysis -- PCA principal component analysis -- LDA linear discriminant analysis -- SSS small sample size problem -- WLDA worst-case linear discriminant analysis -- MMDA max-min distance analysis -- SDP semidefinite programming -- CLMLDA complete large margin linear discriminant analysis -- RMMDA regularized max-min distance analysis -- CLDA Complete LDA -- CCCP constrained concave-convex procedure -- aPAC approximate pairwise accuracy criterion; Dimensionality reduction -- Linear discriminant analysis -- Max-min distance analysis -- Shannon entropy; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.12.030
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Several dimensionality reduction methods based on the max-min idea have been proposed in recent years and can obtain good classification performance. In this paper, inspired by the idea, we develop max-min linear discriminant analysis (MMLDA), which maximizes the minimum ratio of each two-class scatter measure to the within-class scatter measure. However, the method ignores equal emphasis on the distances between class centers and there may be room to improve the classification performance. We then propose regularized max-min linear discriminant analysis (RMMLDA), which introduces the Shannon entropy and the corresponding distance difference regularization terms based on MMLDA. The changing trends of distances between class centers can be precisely controlled in optimization and the separation between classes can be emphasized approximately equally. As a result, RMMLDA may obtain better classification performance. Experiments on synthetic data sets and three publicly available data sets demonstrate its effectiveness. Abstract : Highlights: We propose max-min linear discriminant analysis. We present regularized max-min linear discriminant analysis using Shannon entropy. The improved form can obtain better classification results for high-dimensional data.