Detection of tumors using statistical measures

Classification of tumors is a challenging task in the area of bioinformatics. Gene expression levels are generally used in the diagnosis of tumors. Using thousands of genes to diagnose tumors is a crucial task. Hence a 2-level selection strategy is proposed to identify the most informative genes to diagnose tumors. The most informative genes are found in the first level of selection by three statistical measures. They are variance ratio, prediction strength and T-statistics. The informative genes obtained from these statistical measures are considered as initial members in Differential Evolution algorithm. In the proposed work, the scaling factor is made to vary dynamically to evolve the mutant member in the population. The results obtained with the proposed work prove that it suits better to classify tumor samples.