Akademik Veri Yönetim Sistemi
INFORMATION SCIENCES, sa.741, ss.1-48, 2026 (SCI-Expanded, Scopus)
Feature selection is a prominent step in data preprocessing, and its primary goal is to maximizethe performance of machine learning algorithms by eliminating irrelevant features in the dataset.The fact that the metaheuristics can be easily applied to several types of problems and providepractical solutions can be regarded as an indispensable resource. This study presents a binaryversion of the artificial locust swarm optimization algorithm for selecting the highest qualityfeatures from high-dimensional datasets. Coupled with this, a dynamic time-varying S-shapedtransfer function family based on agent motion is designed to map the continuous search spaceinto a discrete one. The proposed transfer function family draws on the precision function ofALSO that controls the trade-off between exploration and exploitation and solution sensitivity.The performance of the algorithm was gauged using the KNN classifier on 32 high-dimensionaldatasets selected from various application fields. Experimental results prove that the BALSO al-gorithm outperforms state-of-the-art algorithms pertaining to various evaluation measures.Moreover, the superiority of the algorithm was emphasized from a statistical perspective.Consequently, the dynamic time-varying BALSO algorithm based on the motion of the searchagent proves to be a promising method concerning both classification and optimization abilities.