<p>With the intensification of the global climate crisis, attaining the dual-carbon goals has emerged as a significant strategy for China to cope with climate change. However, existing studies still failed to provide a transparent and efficient approach for analyzing carbon emission reduction paths. Hence, this study provides a novel approach from the perspectives of carbon emission prediction and efficiency evaluation. Specifically, the approach is based on interpretable machine learning (ML), including Shapley additive explanation (SHAP) and cumulative belief rule base (CBRB), to achieve carbon emission prediction firstly, and then the data envelopment analysis (DEA) is used to analyze the management efficiency of the predicted carbon emissions. The results of empirical analysis show that: 1) The approach has ability of interpretability to effectively analyze the contribution degree of indicators on reducing carbon emissions; 2) The use of CBRB can effectively enhance the accuracy and efficiency of the approach on carbon emission prediction; 3) By integrating the efficiency evaluation of carbon emissions, it is possible for the approach to further provide a reference basis for the formulation of carbon emission policies. Additionally, this research also can provide insightful experiences for carbon emission reduction paths, which is conducive to proposing targeted carbon neutrality policies and management measures, thereby promoting the transformation of various regions in China towards a low-carbon development path.</p>