Machine learning-driven analysis of carbon emission heterogeneity and characteristics in the final year of COVID-19 control measures: unsupervised clustering insights from listed companies in China

Abstract

This study investigates the heterogeneity of carbon emissions during the final year of COVID-19 control measures, employing advanced machine learning techniques to reveal intricate patterns often overlooked by conventional approaches. By combining descriptive statistics with clustering methods such as DBSCAN, Gaussian Mixture Models (GMM), and K-Means, the research uncovers latent structures in emission data, effectively addressing outliers and refining the understanding of distribution dynamics. Key findings demonstrate notable variability across different enterprise clusters: low-emission enterprises exhibit stable emissions with sporadic high-value anomalies; moderate-emission enterprises show pollutant fluctuations; and high-emission enterprises present considerable variation in greenhouse gas outputs. Additionally, enterprises under unique circumstances display distinctive emission profiles, shaped by exceptional operational conditions. The growth of emerging sectors and changes in production processes during the pandemic contributed to an increase in carbon emissions and wastewater discharge. These insights highlight the necessity of tailored interventions to address the specific challenges of each cluster, thereby improving the reliability and interpretability of emission data. The novel application of machine learning techniques offers significant support for policymakers, aiding in the design of targeted strategies to mitigate environmental impacts and foster sustainable development. This approach emphasizes the importance of adaptive management and flexible policy responses to diverse industrial contexts. Future research should delve deeper into the dynamic interactions between industrial activities and environmental conditions, further advancing global climate change mitigation efforts.