Application of the AutoGluon AutoML Framework for Cryptocurrency Fraud Detection

With the rapid development of electronic payments, virtual currencies have become increasingly widespread. Consequently, electronic fraud has evolved to become more concealed and sophisticated, traditional detection methods struggle to address new and emerging fraudulent techniques. To address this issue, this paper proposes an efficient method for constructing a fraud detection feature framework based on a real-world transaction fraud dataset. This approach aims to enhance the predictive performance of machine learning models in identifying fraudulent behaviors, and the dataset incorporates features such as transaction timestamps, transaction amounts, and transaction addresses, which represent behavioral patterns commonly associated with fraudulent activities. Furthermore, this paper employs the AutoGluon automated machine learning to determine the optimal model combination by systematically comparing various architectures involving different stacking layers and bagging folds. The experimental results show that the proposed AutoGluon-based framework significantly improves the predictive accuracy and stability of financial fraud detection tasks, and providing robust technical support for mitigating virtual currency fraud risks in the finance era.