Research on Illegal Transaction Detection in Ethereum Network Based on Machine Learning

Abstract

The fast development and growth of blockchain technology and cryptocurrencies, but most importantly, the fast diffusion of Ethereum, opened new chances for financial innovation but aggravated the risks of illegal activities such as money laundering. This paper discusses using machine learning techniques to detect illegal transactions over the Ethereum network. The dataset used is from Kaggle and includes a record of transaction features between Ethereum accounts; it has a high degree of class imbalance. Three machine learning models were used to classify transaction legality: Logistic Regression, Random Forest, and Extreme Gradient Boosting; this is referred to as XGBoost. Class balancing and data preprocessing are ways to improve model performance. The evaluation metrics were chosen as Accuracy and Area Under the Receiver Operating Characteristic Curve (ROC AUC). Experimental results show that the best performance of the XGBoost model was 98.52% in accuracy, while Random Forest was the best on ROC AUC, showing very strong classification capabilities. This work has shown the potentiality of machine learning in the improvement of blockchain security and provided useful lessons that might be applied to the development of scalable AML systems.