Triangular-based sine cosine algorithm for global search and feature selection.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• The sine cosine algorithm (SCA) is a population-based metaheuristic.
• SCA faces challenges in balancing exploration and exploitation, especially in high-dimensional spaces, leading to slow convergence and reduced precision.

Purpose of the Study:

• To introduce a novel enhanced sine cosine algorithm (TTOSCA) with improved optimization capabilities.
• To develop and evaluate a binary variant (BTTOSCA) for discrete optimization problems like feature selection.

Main Methods:

• Proposed TTOSCA by integrating a triangular optimization (TO) strategy and a theft mechanism (TM) into the SCA.
• Evaluated TTOSCA against 27 algorithms on IEEE CEC2017 benchmark functions using Wilcoxon signed-rank test.
• Assessed BTTOSCA on 17 UCI datasets for feature selection, including medical and gene data.

Main Results:

• TTOSCA demonstrated significant improvements in precision and convergence speed compared to existing algorithms.
• BTTOSCA effectively balanced exploration and exploitation in feature selection.
• BTTOSCA achieved smaller feature subsets without sacrificing classification accuracy.

Conclusions:

• TTOSCA offers a robust enhancement to the SCA for continuous optimization.
• BTTOSCA is a powerful and efficient tool for feature selection in high-dimensional and discrete domains.