Constrained Hybrid Metaheuristic: A Universal Framework for Continuous Optimisation

A team of researchers has introduced a new, highly adaptable optimization algorithm called the constrained Hybrid Metaheuristic (cHM). Published on arXiv by Piotr A. Kowalski, Szymon Kucharczyk, and Jacek Mańdziuk, the framework is designed as a universal solver for continuous optimization problems. Unlike many algorithms tailored to specific function types, cHM operates across a broad spectrum, handling complex challenges like non-convexity, non-separability, and varying smoothness. Its key innovation is a modular, two-phase structure that dynamically adapts its search behavior by harnessing synergy between candidate solutions and component metaheuristic strategies, applying the most appropriate technique at each stage of the process.

In an extensive experimental evaluation, cHM was tested against traditional metaheuristics on 28 benchmark functions. The results demonstrated that it consistently matched or outperformed existing methods in terms of both final solution quality and convergence speed. To prove its practical value beyond theoretical benchmarks, the researchers also applied cHM to a real-world feature selection problem within data classification. The successful application underscores the algorithm's potential as a versatile and effective black-box optimizer, suitable for deployment in complex, real-world scenarios where problem properties may be unknown or heterogeneous, bridging the gap between academic research and industrial application.