Ahod Alghuried (University of Central Florida), David Mohaisen (University of Central Florida)
Phishing attacks remain a critical threat to the Ethereum ecosystem, accounting for over 50% of Ethereum-related cybercrimes and prompting the rise of machine learning-based defenses. This paper introduces a comprehensive framework to enhance phishing detection in Ethereum transactions by addressing key challenges in feature selection, class imbalance, model robustness, and algorithm optimization. Through a systematic evaluation of existing approaches, we identify major gaps in practice, particularly in feature manipulation and unsustainable performance gains. Our analytical and empirical assessments demonstrate that the proposed framework improves detection generalizability and effectiveness. These findings underscore the need to refine detection strategies in response to increasingly sophisticated phishing tactics in the blockchain domain.