Abstract
Manual parameter tuning of cyber-physical systems is a common practice, butit is labor-intensive. Bayesian Optimization (BO) offers an automatedalternative, yet its black-box nature reduces trust and limits human-BOcollaborative system tuning. Experts struggle to interpret BO recommendationsdue to the lack of explanations. This paper addresses the post-hoc BOexplainability problem for cyber-physical systems. We introduce TNTRules(Tune-No-Tune Rules), a novel algorithm that provides both global and localexplanations for BO recommendations. TNTRules generates actionable rules andvisual graphs, identifying optimal solution bounds and ranges, as well aspotential alternative solutions. Unlike existing explainable AI (XAI) methods,TNTRules is tailored specifically for BO, by encoding uncertainty via avariance pruning technique and hierarchical agglomerative clustering. Amulti-objective optimization approach allows maximizing explanation quality. Weevaluate TNTRules using established XAI metrics (Correctness, Completeness, andCompactness) and compare it against adapted baseline methods. The resultsdemonstrate that TNTRules generates high-fidelity, compact, and completeexplanations, significantly outperforming three baselines on 5 multi-objectivetesting functions and 2 hyperparameter tuning problems.