Random Feature PRIM for Interpretable Illicit Transaction Detection in Cryptocurrency Networks

We apply Random Feature Patient Rule Induction Method (RF-PRIM) to blockchain security, presenting a novel approach for detecting illicit transactions in cryptocurrency networks that balances interpretability with predictive performance. The pseudonymous nature of blockchain transactions creates significant challenges for regulatory compliance and financial crime prevention. While existing blackbox detection methods achieve high accuracy, they lack the interpretability crucial for forensic investigation and regulatory oversight. Our approach combines random feature projections to handle high-dimensional data with Patient Rule Induction Method (PRIM) to generate interpretable classification rules. Evaluation on the Elliptic dataset (over 200,000 Bitcoin transactions with known illicit labels) demonstrates that RFPRIM achieves competitive performance (F1 score of 0.768) compared to traditional methods while producing humanreadable rules describing suspicious activity patterns. Temporal analysis reveals the evolution of illicit behavior patterns throughout the blockchain timeline. This interpretability advantage enables investigators to understand flagged transactions and adapt to emerging fraud patterns, enhancing blockchain security monitoring tools.