Partial Least Squares Based Dimension Reduction with Gene Selection for Tumor Classification

Analyzing gene expression data from DNA microarrays by commonly used classifiers is a hard task, be-cause there are only a few observations but with thousands of measured genes in the data set. Partial least squares based dimension reduction (PLSDR) is superior to handling such high dimensional problem, but irrelevant features will introduce errors into the dimension reduction process and reduce the classification accuracy of learning machines. Here feature selection is applied to filter the data and an algorithm named PLSDR^g is described by integrating PLSDR with gene selection, which can effectively improve classification accuracy of learning machines. Feature selection is performed by the indication of t-statistics scores on standardized probes. Experimental results on seven microarray data sets show that the proposed method PLSDRg is effective and reliable to improve the generalization performance of classifiers.