Personalized Energy Optimization in Smart Homes Using Adaptive Machine Learning Models: A Feature-Driven Approach

Abstract

The increase in demand for efficient energy smart homes has necessitates the personalized optimization strategies to have a reduction in energy consumption while maintaining user comfort. This research develops a Personalized Energy Optimization System using adaptive machine learning models to analyze household energy patterns and predict consumption in real time. Leveraging the Appliances Energy Prediction Dataset from the UCI repository, we applied supervised learning algorithms such as Gradient Boosting, XGBoost, CatBoost, LightGBM, and Random Forest to identify key factors influencing energy use, including occupancy patterns, appliance usage, and environmental conditions. Through feature engineering, normalization, and one-hot encoding, we enhanced model performance and interpretability. Among the evaluated models, LightGBM achieved the highest accuracy (R²: 0.999573, RMSE: 0.013526), outperforming others in predicting energy consumption. The findings offer data-driven insights for dynamic energy management, optimizing household efficiency, and promoting sustainability.