Composition of Feature Relevancy Based Biomarker Gene Selection in Gene Expression Dataset

Abstract

Cancer gene selection plays a prominent work in the area of Bioinformatics. Gene selection methods aim to retain relevant genes and remove redundant genes. This proposed technique deals on gene selection techniques based on information theory. By investigating the information theory based on composition of feature relevancy, we consider that a excellent gene technique method could boost novel classification of the cancer gene data while reducing gene redundancy. Therefore, a modified gene selection technique called Composition of Feature Relevancy (CFR) is carried out. To assess CFR, the experiments are carrying out on five real-world cancer gene expression data sets and three best classifiers (KNN, Support Vector Machine and Random forest). The modified gene selection technique gives best outcome when competing to other recent technique in terms of accuracy and sensitivity in classification.