Beyond Trial and Error: A Ghanaian contribution to metaheuristic classification and its implications

Author

In the fast-evolving fields of artificial intelligence, operations research, and computational intelligence, metaheuristics have become essential tools for solving real-world optimization problems that are otherwise computationally intractable. My recent study, published in the Indian Journal of Science and Technology (Vol. 17, Issue 27, 2024), offers a comprehensive classification and trend analysis of 654 distinct algorithms developed between 1965 and August 2023. This work fills critical gaps left by earlier classification attempts and introduces a structured taxonomy that can guide future algorithm development.

Metaheuristics are high-level algorithmic frameworks that guide heuristics in exploring large and complex search spaces. Unlike traditional heuristics, which are often rigid and domain-specific, metaheuristics are problem-independent and adaptable. They excel at balancing intensification (local exploitation) with diversification (global exploration). Despite their widespread use, a holistic and up-to-date classification reflecting the diversity and metaphorical inspiration of these algorithms was lacking.

Our study employed a systematic descriptive review approach. We conducted exhaustive searches across five major academic databases: Google Scholar, ScienceDirect, Springer, IEEE Xplore, and ResearchGate. For the first objective, we reviewed 148 studies on existing classification frameworks, of which only six met the inclusion criteria. These studies did not adequately cover the full range of classification characteristics. As a result, we proposed a new eleven-level hierarchical classification system with metaphor vs. non-metaphor as the first level of classification.

Regarding this level, metaheuristics were metaphorically categorized as human, sports, maths, music, physics-chemistry, and bio-inspired (including plant, swarm, and evolutionary algorithms). Swarm intelligence included aquatic, flying, and terrestrial animals, as well as microorganisms. The remaining levels include nature-inspired vs. non-nature-inspired, single vs. population, deterministic vs. stochastic, one neighbourhood vs. multi neighbourhood, local search vs. global search, greedy vs. iterative, memory vs. memoryless, static objective vs. dynamic objective, non-hybrid vs. hybrid, and non-parameterized vs. parameterized methods. This framework also integrates algorithmic behavior dimensions. For Objective 2, we examined 1145 studies and filtered 654 metaphor-based metaheuristics based on strict inclusion and exclusion criteria, which were then explicitly classified using the metaphor-based criterion.

Among the 654 algorithms classified, physics-chemistry algorithms were the most prevalent (20%), followed by human-based (18%) and bio metaheuristic methods. The trend analysis revealed periods of stagnation between 1965 and 1992, with innovation accelerating rapidly after 2000 and peaking in 2020 with 68 new algorithms. These findings highlight both the growth and fragmentation of research in the field, emphasizing the need for structured models of knowledge.

Significantly, this study goes beyond categorization. It highlights redundancies in algorithm design and underscores the need for rigorous benchmarking and comparative studies. By mapping the evolution of metaheuristics, the study provides valuable guidance for selecting or designing algorithms in domains such as logistics, wireless ad hoc networks, energy systems, robotics, and national security.

As a Ghanaian researcher, I see this contribution as positioning Africa within the global conversation on computational optimization. It is time for academic institutions, policymakers, and industry stakeholders to support algorithm research through funding, curriculum inclusion, and practical integration. Ghana, with its growing digital infrastructure and skilled intellectual capital, is well-positioned to lead.

In conclusion, metaheuristics are no longer trial-and-error tools. They are scientifically grounded, metaphor-rich constructs. This classification system equips the research community to navigate the increasingly crowded landscape of optimization algorithms. I hope this work will inspire further innovation and critical evaluation of existing methods, ultimately enhancing the power and applicability of metaheuristic-based solutions worldwide.