A hybrid model for optimum gene selection and Classification

Abstract

The present paper addresses the solution for gene subset selection problem with the help of proposed hybrid model for the binary class classification task under supervised learning. As established facts that filter approach is better than wrapper in terms of computational efficiency and time consumption. On the other hand, accuracy of wrapper is higher in comparison to filter. The dilemma faced by the researchers, can be minimised by developing a hybrid model to utilise the efficiency of the filter and accuracy of wrapper together to get the best possible outcome in terms of optimum accuracy. Here the selection of filter is as ReliefF algorithm and wrapper as randomised gene subset selection algorithm. The proposed model is, to first applying an accurate filter, to find out most relevant gene and further applying the efficient wrapper to extract most informative gene from the previous one and optimum gene subset is appreciated by linear as well as an ensemble classifier to classify public domain viz. leukaemia, ovarian cancer, lymphomas and prostate tumour datasets with fivefold cross validation calculation and also signifies the performance measures of the classifiers along with time measurement. The obtained results justify the applicability of the proposed method. Copyright © 2015 Inderscience Enterprises Ltd.