A data-driven framework for energy efficiency evaluation and load management in IoT-enabled smart grids
Energy efficiency assessment is fundamental to the reliable and sustainable operation of modern smart grids, particularly with the increasing deployment of Internet of Things (IoT) technologies and renewable energy resources. This study presents a data-driven framework for Energy Efficiency Score prediction that integrates Extreme Gradient Boosting Regression (XGBR) and Histogram-based Gradient Boosting Regression (HGBR) with Metaheuristic Optimization Algorithm (MOA), Particle Swarm Optimization (PSO), and Harris Hawks-inspired Learning Optimization Algorithm (HLOA) to optimize model parameters and improve predictive performance. Principal Component Analysis (PCA) is employed to reduce feature redundancy, while time-aware validation and cross-validation are used to improve the reliability of model evaluation. SHAP analysis is further incorporated to quantify feature contributions and improve model interpretability. The proposed framework achieved an R² of 0.9971 and an RMSE of 0.8744, demonstrating highly accurate estimation of the Energy Efficiency Score. The optimized hybrid configurations outperformed their corresponding baseline boosting models, confirming the benefit of metaheuristic parameter optimization for structured smart-grid data. The SHAP analysis further identified grid frequency, environmental conditions, reactive power, active power, and power consumption as among the most influential predictive variables. The results demonstrate that the proposed framework can provide an accurate and interpretable data-driven intelligence layer for energy-efficiency assessment and decision support in IoT-enabled smart grids. The framework therefore offers a scalable approach for predictive energy analytics while complementing, rather than replacing, physics-based grid optimization and operational models.