Hybrid Harris Hawks–Whale Optimization (HHWO): A Novel Bio-Inspired Metaheuristic for Global Optimization with Application to Deep Learning Hyperparameter Tuning
Year: 2026
Type: article
Abstract: Harris Hawks Optimization is an aggressive method, but can become trapped on flat regions because of the method used for exploitation, which is besiege. However, the Whale Optimization Algorithm (WOA) uses a logarithmic spiral for its exploitation phase which results in a slower convergence but retained search diversity. We merge the two into Hybrid Harris Hawks–Whale Optimization (HHWO): The escape-energy value of each agent is still used to trigger exploration/exploitation switching, but as soon as an agent switches into exploitation mode, it takes an HHO step or a WOA spiral step, and an occasional Levy-flight jump (10%) is added on top to prevent the swarm from finding premature equilibrium, and elitism maintains the best solution across generations. We have evaluated 13 standard benchmark functions (five unimodal, five multimodal and three fixed dimension) using five baselines: HHWO, WOA, GWO, PSO, and a real-coded GA, repeating the experiments 30 times for each function and performing Wilcoxon rank-sum significance testing. The results obtained from HHWO show that it achieves the best average rank of the six algorithms (1.92/6), outperforms HHO and GWO on 11 out of 13 functions with statistical significance and is competitive with WOA and GA on the others, which is in line with the No Free Lunch theorem. To illustrate the transferability to a realistic optimization problem, a case-study is summarized on an attention-based CNN for the classification of brain tumors in MRIs, detailed in a companion manuscript.
Source: Computer and decision making.
Authors: Kavita Ghuge, Sandeep Musale, Supriya Mangale
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7225880140
