FS-SVM: A Synergistic Approach to High-Dimensional Feature Selection

Supervised classification in high-dimensional gene expression datasets is frequently used in bio-informatics studies. Feature selection plays an important role in such type of classification problems to avoid over-fitting and to develop a more reliable classifier for the problem at hand. This paper proposes a novel feature / gene selection method, calledWeighted Fisher Score (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">W_FS</i>) to identify the most informative features of gene expression values. The proposed method is based on the Support Vector Machine (SVM) and Weighted Fisher Score to select deferentially expressed features with good classification performance. The performance of the proposed methodWeighted Fisher Score (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">W_FS</i>) is tested on seven openly accessible gene expression datasets, using Random Forest, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i>-Nearest Neighbor and Support Vector Machine classifiers compared with four well-known feature selection methods, i.e. Significant Feature Selection (SigF), Wilcoxon Rank Sum (Wilcox), Maximum relevance, and minimum redundancy (mRmR) and Proportional Overlapping Score (POS). Moreover, the performance has also been assessed by constructing Box-plots, Bar charts and Line charts of the results. The results show that the proposed method Weighted Fisher Score (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">W_FS</i>) stands apart in terms of classification performance.