Improved bald eagle search algorithm for global optimization and feature selection

Abstract

The use of metaheuristics is one of the most encouraging methodologies for taking care of real-life problems. Bald eagle search (BES) algorithm is the latest swarm-intelligence metaheuristic algorithm inspired by the intelligent hunting behavior of bald eagles. In recent research works, BES algorithm has performed reasonably well over a wide range of application areas such as chemical engineering, environmental science, physics and astronomy, structural modeling, global optimization, engineering design, energy efficiency, etc. However, it still lacks adequate searching efficiency and has a tendency to stuck in local optima which affects the final outcome. This paper introduces a modified BES (mBES) algorithm that removes the shortcomings of the original BES algorithm by incorporating three improvements; Opposition-based learning (OBL), Chaotic Local Search (CLS), and Transition & Pharsor operators. OBL is embedded in different phases of the standard BES viz. initial population, selecting, searching in space, and swooping phases to update the positions of individual solutions to strengthen exploration, CLS is used to enhance the position of the best agent which will lead to enhancing the positions of all individuals, and Transition & Pharsor operators help to provide sufficient exploration–exploitation trade-off. The efficiency of the mBES algorithm is initially evaluated with 29 CEC2017 and 10 CEC2020 global optimization benchmark functions. In addition, the practicality of the mBES is tested with a real-world feature selection problem and five engineering design problems. Results of the mBES algorithm are compared against a number of classical metaheuristic algorithms using statistical metrics, convergence analysis, box plots, and the Wilcoxon rank sum test. In the case of composite CEC2017 test functions F21-F30, mBES wins against compared algorithms in 70% test cases, whereas for the rest of the test functions, it generates good results in 65% cases. The proposed mBES produces best performance in 55% of the CEC2020 test functions, whereas for the rest of the 45% test cases, it generated competitive results. On the other hand, for five engineering design problems, the mBES is the best among all compared algorithms. In the case of the feature selection problem, the mBES also showed competitiveness with the compared algorithms. Results and observations for all tested optimization problems show the superiority and robustness of the proposed mBES over the baseline metaheuristics. It can be safely concluded that the improvements suggested in the mBES are proved to be effective making it competitive enough to solve a variety of optimization problems.