A systematic review of explainability in computational intelligence for optimization

This systematic review explores the need for explainability in computational intelligence methods for optimization, such as metaheuristic optimizers, including evolutionary algorithms and swarm intelligence. The work focuses on four aspects: (1) the contribution of Explainable AI (XAI) methods to interpreting metaheuristic performance; (2) the influence of problem features on search behavior and explainability; (3) the role of mathematical theory in providing transparent explanations; and (4) the potential of metaheuristics to enhance the explainability of AI models, such as machine learning (ML). XAI methods such as SHAP, LIME, and visualization techniques provide valuable insights into metaheuristic performance, while landscape analysis and quality diversity approaches reveal algorithm performance across different problem landscapes. The review also explores how metaheuristic algorithms can enhance the interpretability of ML models, turning black-box models into more transparent systems. The work moves on to proposing "Explainergy," a novel concept for integrating explainability into metaheuristic algorithms within the energy domain, enhancing the transparency and usability of optimization models. This review is a foundation for future research combining explainability with evolutionary computation and metaheuristic optimization to address real-world challenges in diverse fields, including energy systems.