Research on carbon drivers of agricultural exports combined with XGBoost and SHAP

Accurate identification of agricultural export carbon emission drivers is essential for developing effective mitigation strategies under global climate change. This study proposes an integrated analytical framework combining the XGBoost machine learning model with SHAP interpretability analysis to capture nonlinear relationships and quantify factor contributions. Using panel data from Central China (1993–2019), model parameters were optimized through grid search and cross-validation. Results show that the optimized XGBoost model achieves high predictive accuracy ( R 2 = 0.935). The SHAP-based attribution analysis quantitatively assessed factor contributions, revealing rural electricity consumption, total agricultural machinery power, and gross agricultural output value as the three most significant positive drivers. These factors displayed mean absolute SHAP values of 0.176, 0.158, and 0.144 correspondingly, whereas total export trade exhibited negative influence. The findings indicate that energy usage and agricultural production scale represent fundamental drivers of agricultural export carbon emission growth, while simultaneously uncovering nonlinear relationships and notable interaction effects between critical factors and emission outputs. Methodologically, this research advances the integration of explainable artificial intelligence within environmental economics, while practically it delivers scientific evidence and policy support for developing differentiated, precise agricultural carbon reduction strategies, contributing significant theoretical and practical implications for advancing agricultural trade’s green transformation.