Processing Bio-medical Data with Class-Dependent Feature Selection

In this paper, we show how to select different feature subsets for different classes, i. e., class-dependent feature subsets, for biomedical data. A feature importance ranking measure, i. e., class separability measure, is used to rank features for each class and obtain class-dependent feature importance ranking. Then several feature subsets for each class are formed and an " optimal" one for each class is determined through a classifier, e. g., the support vector machine (SVM). Our method of class-dependent feature selection is applied on several biomedical data sets and compared with class-independent feature selection. The experimental result shows that our approach to class-dependent feature selection is efficient in reducing feature dimension and producing satisfactory classification accuracy.