Gene Selection and PSO-BP Classifier Encoding a Prior Information

Selecting a relevant and discriminative combination of genes for cancer classification and building high-performing classifier are common and critical tasks in cancer classification problems. In this paper, a new approach is proposed to address the two issues at the same time. In details, BP neural network is employed to construct a classifier, and PSO algorithm is used to select a discriminative combination of genes and optimize the BP classifier accordingly. Besides, sample’s prior information is encoded into PSO algorithm for better performance. The proposed approach is validated on the leukemia data set. The experimental results show that our novel method selects fewer discriminative genes while has comparable performance to the traditional classification approaches.