Machine learning for data verification in emissions trading system

Abstract

Data quality is the cornerstone of any emissions trading system (ETS), although developing an effective assurance mechanism is a considerable challenge. To evaluate potential data quality issues of regulated firms and develop a cost-efficient data verification scheme for the authorities, this study uses domain knowledge and data-driven approaches to identify firms with high data quality risks. Using a unique dataset from China's national ETS, each sample obtains an ensemble outlier score generated by several supervised and unsupervised machine learning techniques, and limited inspection resources are allocated to the facilities with higher scores. Our results show that the models make good predictions where potential misreports are found among the predicted high-risk samples, and 70 % of tampered datapoints are detected in the robust test. The method presented here helps in efficiently verifying firms’ self-report emissions and proposes a feasible solution for intelligent data quality management under ETS context.