Improved Binary Particle Swarm optimization with Evolutionary Population Dynamic for Key Oncogene Selection

Cancer is one of the most deadly diseases in the world, and researchers have been investigating various methods to detect it in the early stages in order to improve the possibility of survival. Unfortunately, cancer data is often very expensive to collect and high-dimensional, where redundant and/or irrelevant features result in the challenges for cancer detection and hide information from useful features or key oncogenes. Therefore, an efficient and effective feature selection algorithm is proposed in this paper to deal with these problems, which based on the classic Particle Swarm optimization (PSO) algorithm, one of the most widely used Evolutionary Computation (EC) techniques. In this paper, the Evolutionary Population Dynamics (EPD) strategy is integrated into PSO to address its limitations and improve its performance on addressing feature selection problems. The proposed algorithm is examined on eight cancer datasets of varying difficulty. Comparisons have been done among the two versions of the proposed approaches, standard binary PSO, and three other feature selection methods in terms of the classification performance, the number of selected features and the convergence behaviors. The results show that in most cases, the proposed EPD mechanism can help PSO to achieve better performance over the compared methods.