Feature Selection and Partial Least Squares Based Dimension Reduction for Tumor Classification

Partial least squares (PLS) is one of the widely used dimension reduction methods for analysis of gene expression microarray data, it represents the data in a low dimensional space through linear transformation, the size of the reduced space by PLS is critical to generalization performance of classifiers. The previous works always determined the top fixed number of components or the top several components by cross-validation. Here we demonstrate the usage of feature selection for PLS based dimension reduction. As a case study, PLS is combined with two feature selection methods (genetic algorithm and sequential backward floating selection) to get more robust and efficient dimensional space, and then the constructed data from the selected components is used as input for the support vector machine (SVM) classifier. We use the method for tumor classification on gene microarray data, experimental results illustrate that our proposed framework is effective both to reduce classification error rates and get compact dimensional space.