Forecast Residential Power Usage via Machine Learning and Climatic Feature Incorporations

Accurate forecasting of residential electricity demand requires models that capture both usage patterns and external environmental influences. This study develops a machine learning–based framework that integrates historical household energy consumption with weather variables to improve prediction accuracy. The UCI Household Electric Power Consumption dataset was augmented with external temperature, humidity, and related climatic parameters to incorporate environmental context. After comprehensive preprocessing and feature engineering, five models—Linear Regression, Random Forest, XGBoost, LSTM, and GRU—were evaluated using MAE, RMSE, and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. Results show that weather-integrated features significantly enhanced forecasting performance, with XGBoost achieving the best accuracy (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> = 0.63). This work demonstrates that incorporating weather conditions into consumption modelling provides more reliable forecasts and offers a scalable approach for smart energy management in residential settings.