A Multi-population Hierarchical Differential Evolution for Feature Selection

In order to address the curse of dimensionality and reduce data processing costs, feature selection (FS) is an essential component. Since the process of selecting features is an optimization problem and prone to getting trapped in local optima, FS methods based on evolutionary computation (EC) can effectively tackle such problems. Therefore, this paper proposes a multi-population hierarchical differential evolution (MPDE) to solve the FS problem. In MPDE, inspired by the chicken swarm optimization (CSO) algorithm, a new multi-population communication mechanism is introduced, where individuals in each sub-population are hierarchically divided, and different levels of individuals adopt corresponding individual enhancement strategies to improve algorithm performance. MPDE is compared with five advanced EC-based FS methods on 18 datasets. The experiments demonstrate that MPDE can achieve higher classification accuracy while selecting fewer features, thus proving its effectiveness in solving the FS problem.