Feature selection using binary particle swarm optimization with time varying inertia weight strategies

In this paper, a feature selection approach that based on Binary Particle Swarm Optimization (PSO) with time varying inertia weight strategies is proposed. Feature Selection is an important preprocessing technique that aims to enhance the learning algorithm (e.g., classification) by improving its performance or reducing the processing time or both of them. Searching for the best feature set is a challenging problem in feature selection process, metaheuristics algorithms have proved a good performance in finding the (near) optimal solution for this problem. PSO algorithm is considered a primary Swarm Intelligence technique that showed a good performance in solving different optimization problems. A key component that highly affect the performance of PSO is the updating strategy of the inertia weight that controls the balance between exploration and exploitation. This paper studies the effect of different time varying inertia weight updating strategies on the performance of BPSO in tackling feature selection problem. To assess the performance of the proposed approach, 18 standard UCI datasets were used. The proposed approach is compared with well regarded metaheuristics based feature selection approaches, and the results proved the superiority of the proposed approach.