Binary Dragonfly Algorithm and Fisher Score Based Hybrid Feature Selection Adopting a Novel Fitness Function Applied to Microarray Data

Abstract

Microarray gene data comprises of a large number of genes and fewer samples. Feature Selection (FS) is performed to select disease-causing genes and enhance the performance of the learning model. FS algorithms can either employ a learning model or use only data information to select the features. Each of these has its own drawbacks. In this paper, we propose a hybrid method that incorporates the advantages of both these aspects to select genes. We also employ evolutionary Binary Dragonfly Algorithm (BDA) for searching an informative subset of features and Radial Basis Function Neural Network (RBFNN) as a learning model. We propose a novel fitness function that helps in the effective selection of the features in BDA. The proposed method is applied to microarray datasets, the results of which is found to be promising.