Abstract
Feature selection is an important component of data mining to reduce the data dimensionality. The feature selection method using swarm-based algorithms such as PSO has obtained satisfactory results. However, its shortcoming such as premature convergence especially in high dimension problems often reduces the performance. In this paper, a hybrid feature selection approach based on rough sets and a new discrete particle swarm optimization (DPSO) has been proposed and investigated in order to solve the above problem. DPSO uses only one random number and three predetermined parameters to update each of the particle’s position which needs less memory allocation for each particle. Experimentation is carried out, using UCI datasets, which compares the proposed method with conventional PSO. The results show that feature subset proposed by DPSO-rough set gives better representation of data and contributes to the improvement of the classification performance.