Machine Learning-Based Detection and Analysis of Suspicious Activities in Bitcoin Wallet Transactions in the USA
This paper addresses the challenge of detecting and tracing suspicious activities in U.S. Bitcoin wallet transactions. Methodologically, it proposes a localized machine learning detection framework that integrates multi-source features—including transaction amounts, timestamps, network flow topology, and address graph structures—and systematically evaluates logistic regression, SVM, and random forest classifiers. A key contribution is the empirical discovery of a strong correlation between unspent transaction outputs (UTXOs) and final wallet balances, alongside validation that random forest significantly outperforms other models in capturing nonlinear fraud patterns. Experimental results demonstrate that the proposed model achieves the highest F1-score, effectively identifying diverse illicit behaviors such as money laundering, coin mixing, and anomalous address clustering. Moreover, it supports real-time detection and regulatory response, offering a practical, deployable technical pathway for cryptocurrency compliance monitoring.