The rise of illegal activities involving blockchain digital currencies is a growing concern. Criminals exploit the anonymity and decentralization of blockchain to increase the accessibility of money laundering, fraud, and illegal fund flows. This challenges the traditional regulatory methods and existing level of security. In this study, financial risk control is combined with machine learning to identify and predict user default risks for preventing illicit activities by users with poor credit. We build a fusion model using LightGBM and XGBoost to analyze 18-month user borrowing, payment, and repayment data for predicting credit default probabilities. The experimental results demonstrate that our approach exhibits high performance in user financial credit analysis, with an AUC, F1-score, and an overall score of 96.8%, 94.7%, and 79.9%, respectively. The identification of low-credit users provides crucial insights for blockchain regulators, thus aiding in the early intervention and prevention of the misuse of digital currencies and ensuring financial security in the blockchain system.