Hybrid information gain based fuzzy roughset feature selection in cancer microarray data

The main objective of this paper is to remove the redundant genes present in the samples and thereby increase the classifier accuracy. This is accomplished by devising a hybrid approach for feature selection that selects a subset of genes from the raw dataset and then classifies the samples based on the training imparted. In this paper, a rank based information gain filter is used for dimensionality reduction. Fuzzy rough set and genetic algorithm methods were combined to form a hybrid approach that selects the prominent genes and removes the redundant ones. The process of classification is performed using extreme learning machines classifier with ten-fold cross validation strategy on three multi class cancer microarray gene expression datasets. Experimental results reveal that the proposed hybrid method produces improved higher accuracy in all the three benchmarked multi-class cancer gene expression datasets obtained from the biomedical repository compared to other techniques available in literature using extreme learning machines classifier.