Abstract:

 In order to improve the recognition rate of colon cancer sample by selecting the related genes, sequential floating search method(SFSM) basing on Chernoff distance was proposed. Every gene was evaluated and selected by analyzing the data set of the colon cancer gene expression profiles. Some candidate feature gene subsets were obtained by searching the selected gene subset with the method of SFSM whose evaluation function was Chernoff distance. Three different classifies, support vector machines, K-nearest neighbors, and RBF neural networks, were used to validate the classified efficiency. The experimental results showed that when β=025, the feature gene combination obtained by SFSM with Chernoff distance as its evaluation function was optimal, and colon cancer sample could be recognized best.

Key words: feature selection, Chernoff distance, sequential floating search method(SFSM), support vector machine(SVM), K-nearest neighbor(KNN), radical basis function neural network (RBFNN)