Residential load forecast: An enhanced machine learning model with socio-economic data and synthetic features

Accurate forecasting of residential electricity loads is a critical step toward enabling demand-side flexibility and supporting efficient and reliable power system operation. This paper presents a novel Feature-Enhanced Residential Energy Forecast (FEREF) model for residential households that does not rely on historical load profiles, which are often unavailable. Instead, the proposed method bases its estimations on: (i) socio-economic data and building characteristics, (ii) synthetic features summarizing past energy behavior, and (iii) weather information. Following a detailed discussion on training data selection and synthetic feature design, the development and tuning of the proposed FEREF model, based on an Extra-Trees regressor, are described. The accuracy and sensitivity of the model are evaluated through representative case studies and benchmarked against alternative approaches from the literature. Results demonstrate that the FEREF model attains an R 2 score of 0.82 and an average MAPE of 37.69%, demonstrating significant performance improvements with respect to the current state of the art, notably achieved while not requiring complete knowledge of historical load profiles. This makes the approach particularly suitable for newly connected households and data-scarce environments.