A Review of Machine Learning Applications for Energy Consumption Forecasting in Schools

Resumen

As schools face growing energy demand under constrained budgets, accurate energy forecasting using Machine Learning has become crucial for improving efficiency and planning targeted energy management strategies. This review examines studies that apply ML techniques in forecasting school and campus energy demands. Based on the methodology of these works, a generalized forecasting framework is proposed, which detailedly outlines: data preprocessing, feature selection, model selection, and implementation of results to guide implementation. Across studies, historical load, weather conditions, occupancy and building attributes are among the most reliable predictors of energy demand. Advanced models such as hybrid LSTM architectures or ensemble approaches generally achieve higher accuracy but require a larger complete dataset, increased computational costs and intensive hyperparameter tuning which limits their feasibility in resource-limited school settings. Simpler and more interpretable alternatives such as MLR often offer sufficient accuracy for schools with limited data availability and resources. Future studies should focus on addressing existing gaps by ensuring transparency and consistency in data and methodological reporting.