Prediction of Household Carbon Emissions Based on SP-LIME and Ensemble Learning Models

In recent years, with the increasing efforts of the industrial carbon reduction policies, the correlation between carbon emissions behind household consumption and household behavior needs more attention. Household Carbon Emissions (HCEs) were evaluated by the combing method of the Multi-Regional Input and Output (MRIO) model and the micro dataset of China Family Panel Studies (CFPS). By coupling the relationship between characteristics and household carbon emissions, this paper puts forward the HCEs prediction method of the ensemble learning, compared with the prediction accuracy of the Decision Tree (DT) models. Based on the framework of behavioral economics, the results apply Submodular Pick-Local Interpretable Model-Agnostic Explanations (SP-LIME) to explain machine learning and explore the oversampling problem of imbalanced data. In addition, the conclusion of this paper shows that the prediction accuracy of this ensemble method as the Lasso-XGBoost model is 90.5S% and 91.23% for urban and rural households, which contributes to HCEs simulation and reduction behavior in the future.