Household-Level Carbon Footprint Forecasting in Nigeria: A Machine Learning Approach with Prediction Error Risk Assessment for Net-Zero Emissions

The study develops a data-driven framework for predicting household CO₂ emissions within a developing economic setting using Talba Estate in Minna, Niger State, Nigeria, as a case study. Hourly data were collected from 10 households for the whole of 2023, encapsulating electricity consumption, income, household size, and climatic parameters. Four machine learning models were benchmarked and evaluated within a prediction-uncertainty risk assessment framework, which quantifies the likelihood and impact of model-based prediction errors rather than policy or environmental risks. The models were trained on a 70/30 train-test split and evaluated within a novel prediction-error risk assessment framework that quantifies model uncertainty. The XGBoost achieved the highest in predictive accuracy among the four, with minimum error rates: MAPE = 0.0073, RMSE = 0.1463, and MAE = 0.0340, an R² of 0.9999, almost a perfect fit. The robustness of the model was also tested by prediction-error risk scoring, with values averaging around zero and stability values at about 0.100 across households. The key innovation is the integration of machine learning forecasting with a structured prediction-error risk assessment framework, applied to high-resolution household data in a resource-constrained setting, a combination rarely addressed in existing literature. The results point toward a promising outlook for hybridizing an advanced machine-learning toolkit with prediction-uncertainty risk quantification toward accurate carbon forecasting in resource constraint context. The findings offer actionable insights for policymakers supporting Sustainable Development Goal 13 and Nigeria’s net-zero emissions targets, advancing scalable carbon monitoring frameworks for developing regions.