Interpretable machine learning analysis of industrial carbon emissions efficiency in Chinese cities

Understanding the spatiotemporal evolution of industrial carbon emission efficiency (ICEE), identifying its key determinants, and developing targeted countermeasures are crucial for improving ICEE and achieving sustainable development and dual-carbon objectives. This study aims to evaluate the ICEE of 285 Chinese cities from 2000 to 2021, exploring its dynamic patterns and key driving factors. We employ a super Slacks-Based Measure (SBM) with nonexpected outputs to calculate ICEE, and utilize the Malmquist Index to decompose dynamic changes into efficiency and technical components. A spatially constrained multivariate clustering analysis is applied to classify cities into six clusters, followed by GeoSHAP (Geospatial Shapley Additive Explanations) to assess the influence of various factors on ICEE. The main findings indicate that: (1) ICEE generally increased over time, with declines and notable fluctuations around 2008 and 2012, mainly driven by technical changes; (2) ICEE displays clear geospatial clustering, with pronounced high- and low-efficiency regions; (3) high-efficiency regions driven by science, technology, and innovation should maintain efficient economic development, whereas low-efficiency regions need to improve energy use efficiency and promote clean energy adoption. These results provide practical guidance for reducing industrial carbon emissions in China and support the effective implementation of national dual-carbon strategies.