Maximum Predictive- Minimum Redundancy gene subset selection from Microarray Data. Predictive Accuracies with Cart and Random Forests Madrid Salford-Systems Data Mining workshop. May 2005

Microarray data analysis is concerned with the extraction of valuable informa-tion from large data sets arranged in hundreds or thousands of columns (genes) and a few number of rows. This work is focused on the classication of normal and tumor tissues from microarray gene expression data. A well known approach to face up the classication is the selection of a subset of genes with high predictive accuracy. We discuss a maximum predictive and minimum redundancy selection procedure that will lead to a group of genes with low correlations and high predic-tive strength. We will test the accuracies of the selected genes, when used in the classi cation of tissue samples, with Cart and Random Forests