Abstract
In recent decades, feature selection has become an important issue since it is capable of reducing the noise and find discriminate features while dealing with high dimensional data. Nowadays, feature selection has been widely applied to many research field, such as medical research related to cancer. Feature selection can also called gene selection in the context of finding discriminate genes related to cancer. Via gene selection, medical personnel can detect the symptoms of cancer at medical early stage. Generally, while dealing with gene selection problem, the maximization of the classification accuracy and the minimization of number of selected genes need to be satisfied simultaneously. As a result, to be closer to reality, in this research, we develop a gene selection method with multi-objective design while dealing with 10 benchmark gene expression datasets. The proposed method is called hybrid filter models and multi-objective simplified swarm optimization (HF-MOSSO). In the HF-MOSSO, firstly, the AHP structure with five filter models is constructed to remove most irrelevant genes. After irrelevant genes have been removed, we develop a multi-objective simplified swarm optimization (MOSSO) using SVM with leave-one-out cross validation (LOOCV) as evaluator to evaluate the performance of gene subsets. In the experimental results, the proposed HF-MOSSO is compared with some previous methods in the literature and the results showed that the proposed method can obtain better accuracy while choosing less genes.