Comparative Evaluation of Machine Learning Methods for Predicting Energy Consumption in Buildings

Accurate forecasting of building energy consumption is vital for sustainable energy management. In this study, we propose a machine learning framework that utilizes an extensive dataset from over one hundred Southern California buildings (January 2018-January 2024) comprising hourly measurements of energy usage, environmental conditions, and operational metrics. By comparing a Decision Tree model with ensemble methods-including Random Forest, Gradient Boosting, XGBoost, and LightGBM-we demonstrate that ensemble approaches achieve superior predictive performance (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2} &gt;0.95$</tex>) and effectively capture seasonal and operational variations. These findings highlight the potential of data-driven techniques in optimizing energy management and informing targeted efficiency interventions in the built environment.