Intelligent Machine Learning Techniques for Energy Consumption Forecasting in Smart Buildings—A Review

Abstract

Accurate forecasting of energy consumption in smart buildings is an important part of environmentally sustainable energy management and smart grid operations. Numerous studies have employed singular Machine Learning (ML) techniques to estimate a building's energy requirements; however, most reviews examine only a limited number of algorithms, forecasting horizons, or datasets. They do not consider how smart homes operate as a whole, how ensemble learning functions, or how to evaluate models. This paper provides a structured and systematic analysis of machine learning-based energy consumption forecasting methodologies within the extensive framework of smart home systems. This review differs from earlier surveys in that it examines (i) the architectural features of smart homes that affect data generation and forecasting accuracy, (ii) a wide range of supervised and ensemble regression methods, such as Random Forest (RF), Gradient Boosting (GB), Extreme Gradient Boosting (XGBoost), Ridge, Lasso, voting regressors, and stacking regressors, and (iii) various evaluation and interpretability frameworks used to assess predictive reliability. The study meticulously assesses algorithms based on their precision, scalability, interpretability, computational cost, robustness to noise, and suitability for nonlinear energy patterns. This review also stresses the utility of hybrid and ensemble machine learning methods for improving prediction accuracy and system stability in changing home environments. The paper uses quantitative metrics, visualization tools, and Explainable Artificial Intelligence (XAI) methods from the literature to compare performance and examine forecasting outcomes. This work provides a comprehensive foundation for developing accurate, scalable, and comprehensible energy forecasting models for next-generation smart homes by integrating smart building system architecture, machine learning methodologies, ensemble techniques, and evaluation frameworks into a unified analytical perspective. The findings indicate unresolved issues and propose avenues for further investigation into hybrid modeling and real-time intelligent energy management systems.