A NEW APPROACH FOR FEATURES SELECTION BASED ON BINARY SALP SWARM ALGORITHM

Metaheuristic techniques become considerably popular in solving feature selection (FS) problems due to their flexibility and ability to avoid the local optimum problem. Features selection is important and essential mean to tackle the classification problems through choosing an optimal features subset according to a certain criterion. FS is used to reduce dimensionality and remove noise from data, these are given rise to speed of learning, simplicity of rules, visualizes the data and predictive accuracy. Salp Swarm Algorithm (SSA) is a new metaheuristic algorithm that emulates the inbred behaviour of the Salp chain. In this study, a new FS approach applies the native SSA in machine learning domain to select the optimal feature group on the basis of wrapper mode. Subsequently, SSA is hybridised with a mutation operator. Mutation is embedded to act as an internal operator and consequently maintain diversity and improve the exploration ability within the SSA. The performance of SSA with mutation operator (SSAMUT) on 16 datasets from UCI Machine Learning repository is evaluated and compared with that of the native SSA and other related approaches in the literature. Experimental results proved the efficiency of the proposed approaches in solving search space problems. SSAMUT presents the most excellent performance compared with those of other approaches on all datasets.