Blockchain technology has emerged as a secure and decentralized framework for digital transactions; however, its open and pseudonymous nature also presents significant challenges related to fraudulent activities and malicious nodes. This study investigates the application of machine learning models for blockchain node classification and fraud detection, evaluating three models: Random Forest, XGBoost, and Neural Network. The research leverages a dataset of 10,000 blockchain transactions with 16 attributes, including transaction fees, block scores, stake distribution rates, and coinage. The results demonstrate that machine learning models can effectively classify blockchain nodes with high accuracy. Among the evaluated models, the Neural Network classifier outperformed the others, achieving an accuracy of 95.3%, precision of 95.1%, recall of 95.6%, and an F1-score of 95.3%. Comparatively, XGBoost achieved an accuracy of 94.1%, while Random Forest scored 92.4%. Feature importance analysis highlighted Block Score (0.38), Transaction Fee (ETH) (0.30), and Stake Distribution Rate (0.15) as the most significant factors influencing classification outcomes. Furthermore, confusion matrix analysis revealed that the Neural Network model produced 4780 true positives and 4440 true negatives, with only 200 false positives and 580 false negatives, demonstrating its robustness in identifying fraudulent nodes. Despite these promising results, real-world deployment presents several challenges, including the evolving nature of fraudulent strategies, real-time detection requirements, and scalability concerns. Future research should explore real-time learning techniques, integration of network-based features, decentralized fraud detection mechanisms, and cross-chain anomaly detection to improve model adaptability and effectiveness. By advancing these methods, machine learning-driven fraud detection can contribute to a safer, more transparent, and resilient blockchain ecosystem.
Agung Budi Prasetio (Mon,) studied this question.