An Ensemble Classifier Based on Selective Independent Component Analysis of DNA Microarray Data

Our study tried to deal with a gene expression problem from the view of factor analysis. In order toovercome the instability problem caused by using traditional Independent component analysis (ICA), an ensemble classifier of DNA microarray data based on a selectiveICA method was proposed. At first, we analyzed the reconstruction error of each gene and selected a subset of independent components, which contributed relatively smallreconstruction errors, to reconstruct new samples. Afterthat, several Support vector machine (SVM) sub-classifierswere trained simultaneously. Finally, the best SVM subclassifiers with high correct rates were selected to participate in the ensemble, using a majority voting method. Results on three publicly available DNA microarray datasetsshow the feasibility and validity of our proposed method.