A Meta-heuristic Hybrid Wrapper Method based on Feature Selection for Classification of Biological Samples

Abstract

Cancer is the vital cause of death across the Globe.Microarray technology is regarded as a promisingdiagnostic and classification tool for cancer. Itexplores genetic mutations occurring within acancer cell. Dimensionality reduction techniques(DRT) are vital in microarray based data analysis.The microarray data contains a huge number ofattributes or dimensions, which can adversely affectperformance parameters of the model. Hence it isnecessary to identify most relevant attributes to beretained and discard rest attributes. Statistical andmachine learning (ML) techniques are employed toidentify the majority of important genes or attributesto be retained. Two wrapper hybrid wrapper modelsare proposed for the feature selection purpose. Thefirst hybrid method combines the Grey WolfOptimisation (GWO) with the Jaya optimisationmethod, whereas second hybrid method combinesGWO and Particle swarm optimization (PSO)algorithm. These two hybrid models are appliedindividually on four benchmark microarray datasetscontaining data on cancer of the central nervoussystem, Breast cancer, ovarian cancer andleukaemia cancer to get reduced datasets.Classification algorithms Support Vector Machine(SVM), Decision Tree (DT), Random Forest (RF),Naive Bayes (NV), and Linear Discriminant Analysis(LDA) are classification models used individually toclassify malignant and benign genes from eachcategory of reduced data sets with stratified 10-foldcross-validation. Classification accuracy of allclassifiers on individual dataset for both wrapperhybrid models is compared with each other.