Microarray data feature selection using hybrid genetic algorithm simulated annealing

Abstract

Microarray data feature selection is crucial for the development of a viable cancer diagnostic system based on microarray data. This paper assesses the effectiveness of the Hybrid Genetic Algorithm Simulated Annealing (HGASA) algorithm in selecting features for various classification architectures. HGASA combines the parallel search capability of Genetic Algorithm (GA) with the flexibility of Simulated Annealing (SA). The algorithm is guided by Separability Index, which quantifies the extent of class separability demonstrated by a combination of features. Four classifiers are used in the assessment: Artificial Neural Network (ANN), Support Vector Machine (SVM), Naïve Bayesian Classifier (NBC) and K-Nearest Neighbour (KNN) classifier. Results from HGSA is compared to those from standard GA as well as to those from Population based incremental Learning (PBIL) algorithm. Two data sets are used facilitate this analysis: a prostate cancer data set and a lymphoma data set. For the prostate cancer data set, features selected by the HGASA attained the highest classification accuracy on the SVM classifier with an accuracy of 88%. For the Lymphoma data set, the highest classification accuracy was attained using the ANN classifier, which attained an accuracy of 95%. The performance of the HGASA is ascribed to its ability to search the feature space more thoroughly by employing a deeper exploration of the feature space, when compared to GA and PBIL.