Supporting energy consumption prediction: a sustainable approach

Balancing the utilization of model resources with the accuracy of the prediction is a significant challenge in energy consumption analysis. It is then essential to select models that combine computational efficiency with suitable predictive accuracy to support sustainable environments. We propose an approach to evaluate the sustainability level of popular machine learning (ML) and deep learning (DL) models, offering decision-makers the flexibility to select the most appropriate model, considering resource efficiency and prediction reliability. It is demonstrated using a novel dataset comprising detailed energy consumption data from four major social mass-housing buildings in Rome. It is collected over two years and is enriched with energy bills, resident surveys, and daily usage patterns.