ML Based Assessment of Household Carbon Emission in Nepal

Abstract

This study develops a machine learning (ML)-based framework to assess and mitigate household carbon emissions in Nepal, leveraging a stacking ensemble model (Random Forest + Gradient Boosting meta-regressor) that achieves high predictive accuracy (MSE: 112.17, R2: 0.98). By analyzing data from 4,000 households across energy use, transportation, waste, and dietary habits—collected via a structured Google Forms survey and processed using feature selection and Z-score normalization—the system provides personalized carbon footprints and reduction strategies, validated against IPCC benchmarks. The web-based FastAPI-React tool identifies high-impact factors (e.g., LPG consumption, bottled water usage, rainwater harvesting) and effective mitigation measures (e.g., solar adoption), offering actionable insights for households and policymakers to support Nepal’s climate goals. This work advances scalable, context-aware ML solutions for sustainability in developing regions.