At CASE 25, the authors outline their approach to training a transferable machine‑learning model capable of accurately predicting household consumption patterns in urban, suburban and rural Cypriot settings. Leveraging a dataset of 12 months of smart‑meter readings from 150 homes, they compare random forests, gradient boosting and neural nets, finding that an ensemble of gradient boosting machines yields the lowest mean absolute error (~8 %). Future work will incorporate weather, occupancy and socioeconomic features to further boost generalizability, with the aim of empowering community managers and utilities to plan demand‑side interventions.
M. A. Qureshi, N. Christofides, T. Leontiou and M. Lestas, “Towards a general Machine Learning model for energy profile predictions to serve Smart Communities in Cyprus,” Conference on the Advancements in Sustainable Engineering (CASE 25), Cyprus, 2025.