Memory-based Harris hawk optimization with learning agents: a feature selection approach

Feature selection is a vital pre-processing phase for most machine learning and data mining courses. This article proposes new variants of the Harris hawk optimization called memory energetic Harris hawk optimization (MEHHO1 and MEHHO2) to select the optimal features for classification purposes. The MEHHO approaches adopt an energetic learning strategy and memory saving and updating mechanism. The former extends the chance of the algorithm escaping the local solutions, while the latter boosts the exploitation behavior. The proposed approaches are applied in the feature selection domain for assessing a subset of high discriminative features. The proposed approaches are evaluated on 13 low-dimensional and eight high-dimensional datasets. Also, the proposed approaches are utilized to solve the feature selection problem for the classification of electromyography signals. Our results prove the capability of the proposed approaches to find the optimal feature subset compared to the other five well-known optimization algorithms. Thus, the proposed MEHHO is expected to be a promising and effective technology to solve the feature selection problem.