Chapter 37 - Gene selection for microarray data classification based on mutual information and binary whale optimization algorithm

Abstract

Detection of disease causes can be facilitated by various procedures or techniques. One of the main challenges in classifying gene expression data is the development of an efficient approach to obtain disease information. When used to solve problems involving gene selection, bio-inspired algorithms rank among the best. The Whale Optimization Algorithm (WOA) can converge faster than previous techniques and is less computationally expensive. In this study, we propose a new binary version of WOA using a recent transfer function (TF) called Taper shaped TF. To handle gene selection in microarray data classification, the Binary-WOA (BWOA) approach is combined with mutual information (MI). MI is used as a pre-filtering fitness function in the first step of this two-step method to assess gene relevance and redundancy. Then, the BWOA fitness function, which uses a K-Near Neighbors (KNN) classifier and the number of selected genes, is used to develop subsets of genes. We compared the proposed M-BWOA method with recent state of art algorithms to evaluate its performance. The experiments were performed on sixteen benchmark datasets with different number of classes. The experimental results show that the M-BWOA algorithm provides better classification accuracy.