Mutual information based reduction of data mining dimensionality in gene expression analysis

This article introduces a novel method for reducing dimensional complexity of classification problems which are frequently present in gene microarray analysis. Revealing the most relevant subset of genes among few thousands of analyzed genes is necessary to get accurate disease classification. Attribute (gene) filter was developed for such a purpose. The filter, first introduced as mutual information feature selection (MIPS) was coupled with the support vector machines (SVM) classifier in the leave-one-out (LOO) loop, which resulted in an efficient and reliable tool named MIFS/SVM hybrid. The set of gene microarrays, which consists of two leukemia types, was used as a benchmark. That particular set was thoroughly analyzed by others. Hence, it was appropriate to use it for testing the accuracy of MIFS/SVM hybrid based filter