Detecting Carbon-Credit Laundering Through Integrated ESG and Transaction-Network Analysis

This paper investigates laundering risks in carbon-credit markets by integrating environmental, social, and governance (ESG) data with financial transaction records. A dataset covering 1,247 firms across three jurisdictions from 2018–2023 was assembled, combining carbon-registry information, credit-trading logs, ESG disclosures, and banking records. A hybrid detection model combining a graph neural network with a gradient-boosted classifier was trained using 1,350 labeled suspicious cases and 6,700 normal cases. The model achieved an AUC of 0.92 and a precision of 0.71 at 70% recall, outperforming rule-based systems by 17.3 percentage points. Entities flagged by the model frequently showed abnormal credit recycling within 30 days and emissions inconsistencies exceeding sector benchmarks by more than 25%. These findings indicate that integrated multi-source analysis can effectively identify carbon-credit laundering.