The rise of blockchain technology and cryptocurrencies such as Bitcoin and Ethereum has created new avenues for both lawful and illicit activities, including illegal firearm transactions. This study applies a combination of graph-based analysis, functional data techniques, and machine learning to detect and classify suspicious activities related to firearm trafficking on blockchain networks. A Random Forest model, achieving a precision of 0.907 and recall of 0.786, was used to identify illicit Bitcoin addresses, while a multi-target classifier categorized these addresses by specific types of illicit activity. For Ethereum, an XGBoost model achieved a precision of 0.9864 and an accuracy of 0.9901, demonstrating robust detection of suspicious accounts. Feature engineering and a rule-based system further enhanced model performance, though challenges remain in addressing misclassifications, particularly in distinguishing subtle transaction patterns. These findings underscore the potential of machine learning in blockchain forensics, providing critical insights for law enforcement efforts to combat illegal firearm trading.