A novel feature extraction approach based on ensemble feature selection and modified discriminant independent component analysis for microarray data classification

Microarray data play critical role in cancer classification. However, with respect to the samples scarcity compared to intrinsic high dimensionality, most approaches fail to classify small subset of genes. Feature selection techniques can reduce the dimension of the problem, which can reduce computational cost of the microarray data classification. However, previous studies have shown that feature extraction methods can also be useful in improving the performance of data classification. In this paper, we propose an ensemble schema for cancer diagnosis and classification that has three stages. At first, a hybrid filter-based feature selection method using modified Bayesian logistic regression (BLogReg), Ttest and Fisher ratio is applied for selecting genes. In the second stage, selected genes are mapped via the proposed PSO-dICA method which is a modification of dICA. Finally, mapped features are classified using SVM classifier. To demonstrate the effectiveness of the proposed method, some traditional microarray data including Colon, Lung cancer, DLBCL, SRBCT, Leukemia-ALL and Prostate Tumor datasets are used. Experimental results show the efficiency and effectiveness of the proposed method.