Machine Learning-Based Prediction of Household Energy Consumption in Ghana

Abstract Accurate forecasting of residential electricity demand is a pressing issue in Ghana, where supply constraints and growing consumption continue to challenge the stability of the energy sector. This study develops and evaluates a machine learning framework for predicting household electricity use in the Bono region using real consumption data provided by the Northern Electricity Distribution Company (NEDCO). The dataset, which contains more than 156,000 records on demographic, economic, and appliance-related factors, was analyzed using four models: Gradient Boosting, XGBoost, Random Forest, and Linear Regression. The models were trained and validated using an 80/20 split and five-fold cross-validation. Gradient Boosting achieved the strongest performance (RMSE = 17.75 kWh, R2 = 0.9140), with household size and appliance count identified as the most influential predictors. Beyond model development, a web-based platform was built to provide households, energy managers, and administrators with real-time forecasts and expenditure projections that reflect Ghana's tariff structures. These results demonstrate the value of localized data and machine learning in guiding household budgeting, supporting capacity planning, and informing policy design.