Potential Fraud Detection in Carbon Emission Allowance Markets Using Unsupervised Machine Learning Models

Fraud detection in financial markets is a critical task, particularly in the Carbon Emission Allowance (CEA) market, where fraudulent activities can distort market integrity. This paper explores the application of unsupervised machine learning models for detecting fraudulent activities in the CEA market from 2018 to 2023. We evaluated several models, including Isolation Forest, One-Class SVM, Autoencoder, DBSCAN, LOF, K-Means, Elliptic Envelope, and PCA, using tick-by-tick trading data resampled into daily data. Volume outliers were identified using Z-scores, and their correlation with detected frauds was analyzed. Additionally, we integrated these fraud detection models with a Simple Moving Average Crossover trading strategy to assess their impact on trading performance. The results show that models such as Isolation Forest and Elliptic Envelope significantly improve trading outcomes, with notable increases in net profit and win rate while reducing drawdown. Our findings demonstrate that incorporating fraud detection into trading strategies can enhance market performance by mitigating risks associated with fraudulent activities.