The Impact of the Digital Economy on Carbon Emissions: Evidence From Machine Learning, Graph Neural Networks, and the EKC Hypothesis

Abstract

This study utilizes panel data from 30 Chinese provinces spanning 2007 to 2023 and integrates machine learning and graph neural network (GNN) approaches to examine the spatial dynamics of carbon emissions. It aims to systematically evaluate the impact pathways of the digital economy on carbon intensity and to uncover its spatial diffusion patterns and regional heterogeneity. The empirical findings are threefold. First, the digital economy significantly reduces carbon intensity, consistent with the Environmental Kuznets Curve (EKC) hypothesis, and this effect exhibits clear heterogeneity across economic development levels and regions. Second, due to the existence of spatial spillover effects, GNN models outperform traditional machine learning methods in carbon emission prediction tasks. Third, carbon intensity displays strong temporal inertia and negative spatial spillovers across regions. Notably, spatial diffusion capacity and sensitivity to the digital economy vary substantially: central and western regions exhibit stronger spillover effects, northeastern provinces show more pronounced internal feedback mechanisms, while eastern coastal areas demonstrate relatively weaker effects. Overall, this study expands the analytical perspective on the digital economy's role in carbon mitigation and provides theoretical and empirical support for the design of differentiated emission reduction policies and coordinated regional governance.