Akademik Veri Yönetim Sistemi
6th Middle East and North Africa Communications Conference, MENACOMM 2026, Hammamet, Tunus, 12 - 14 Mayıs 2026, (Tam Metin Bildiri)
Counterfactual explanations have emerged as a practical concept for translating machine learning predictions into actionable recommendations, particularly in smart city applications. In this study, the primary focus is on generating highquality counterfactual explanations using single-objective and multi-objective Particle Swarm Optimization (PSO) methods in order to provide actionable recommendations for reducing nexthour electricity consumption. The proposed approach formulates counterfactual generation as an optimization problem, where proximity, sparsity, and validity are explicitly considered. A Random Forest regression model is employed as the underlying forecasting model for next-hour electricity consumption. The generated counterfactual explanations are evaluated using standard counterfactual quality metrics, including validity, proximity, sparsity, diversity, and plausibility, and are compared against commonly used baseline methods. Experimental results demonstrate that PSO-based approaches consistently yield more realistic and actionable counterfactuals among competing methods. This study highlights the suitability of PSO-driven counterfactual generation as a robust recommendation mechanism for shortterm electricity consumption reduction in smart building environments.