Feature selection strategy based on hybrid horse herd optimization algorithm and perturbation theory: an mRMI approach

In the big-data era, conveying information related to data can sometimes lead to redundancy and irrelevance. This large volume of information not only lacks benefits for optimal decision-making but also increases the costs associated with data collection, storage, and processing. In this view, dimension reduction can be an effective solution. Therefore, we propose a new minimum Redundancy and Maximum Interaction (mRMI) feature selection method. The proposed method controls redundancy through perturbation theory and clustering. It then selects representative features from each cluster using horse herd optimization algorithm (HOA) to ensure that the selected subset of features interacts effectively for classification. We call it PHOAFS (perturbation theory HOA feature selection), which is validated on microarray datasets. The effectiveness of PHOAFS is investigated by comparing it with both conventional and new feature selection algorithms found in the literature. The experimental results show that the proposed method (PHOAFS) can achieve a smaller number of features while maintaining or exceeding the accuracy of other algorithms.