BMPA-TVSinV: A Binary Marine Predators Algorithm using time-varying sine and V-shaped transfer functions for wrapper-based feature selection

The feature selection problem is one of the pre-processing mechanisms to find the optimal subset of features from a dataset. The search space of the problem will exponentially grow when the number of features increases. Hence, the feature selection problem is classified as an NP-hard problem, and exact algorithms cannot find the optimal subset at a reasonable time. As a result, approximate algorithms like meta-heuristic algorithms are extensively applied to solve the problem. The feature selection problem is a discrete (binary) optimization problem; consequently, a discrete meta-algorithm can be employed to find the optimal subset of features. One of the recently introduced meta-heuristic algorithms is Marine Predator Algorithm (MPA), which has shown good solutions to many continuous optimization problems. In this study, a novel Binary Marine Predator Algorithm using Time-Varying Sine and V-shaped transfer functions (BMPA-TVSinV) is proposed to find the optimal subset of features in datasets. The proposed algorithm applies two new time-varying transfer functions to convert the continuous search space to the binary one. These transfer functions considerably improve the performance of BMPA-TVSinV. Several well-known datasets with high-dimensional features and three coronavirus disease (COVID-19) datasets have been selected to compare the results of BMPA-TVSinV with some recently introduced binary meta-heuristic algorithms for the feature selection problem. The results show the superiority of BMPA-TVSinV in achieving high classification accuracy and feature reduction rate. The source code of BMPA-TVSinV for feature selection problem is publicly available at https://www.mathworks.com/matlabcentral/fileexchange/115315-bmpa-tvsinv-a-binary-metaheuristic-for-feature-selection.