Efficient data-driven prediction of household carbon footprint in China with limited features

Household carbon footprints significantly contribute to global carbon emissions, yet the limited features hinder the accurate assessment in regions with inadequate data. In this study, we developed a data-driven model for evaluating household carbon footprint in China based on a nationwide questionnaire survey. Employing explainable machine learning, we identified the most influential features impacting the model outputs, which were utilized to assess household carbon footprints across Chinese provinces and cities from 2006 to 2021. Additionally, we constructed 27 scenarios with varying peak carbon times, technological evolution pathways, and cumulative carbon budgets to analyze climate change mitigation options for policymakers. The results demonstrate that using the XGBoost-TPE method, the model constructed with only five pivotal features achieves a confidence level of 87.4%. From 2006 to 2021, China's provinces exhibited fluctuating growth trends in household carbon footprints, with economically developed and eastern regions demonstrating higher per capita household carbon footprints. Scenario modelling reveals that combining multiple moderate scenarios yields a 26% increase in effectiveness compared to a single scenario. Promoting low-carbon household food, energy, and transport policies earlier or more aggressively can result in an earlier peak in household carbon footprints and a 2.5% greater carbon reduction.