Abstract
The sine cosine algorithm (SCA) is a popular population-based optimization technique that utilizes sine and cosine functions to navigate complex search spaces. However, its inherent simplicity can hinder optimal exploration-exploitation balance, particularly in high-dimensional problems, leading to slow convergence and reduced precision. To address these issues, we propose a novel triangular optimization (TO) strategy combined with a theft mechanism (TM), resulting in an enhanced algorithm named TTOSCA. We rigorously evaluate TTOSCA against 27 competing algorithms using the IEEE CEC2017 benchmark functions and apply the Wilcoxon signed-rank test to assess performance. Our results indicate that TTOSCA significantly improves precision and convergence speed. Furthermore, we develop a binary variant, BTTOSCA, to validate its effectiveness in discrete spaces through feature selection experiments on 17 datasets from UCI, including medical and high-dimensional gene data. Comparative analysis shows that BTTOSCA excels in both exploration and exploitation, achieving smaller feature subsets without compromising classification accuracy. This positions BTTOSCA as a powerful tool for feature selection in high-dimensional datasets.