Beyond Node Attributes: A Controlled Evaluation and Ablation of Local Interaction Statistics for Ethereum Phishing Detection
In briefEthereum phishing detection: testing whether extra account-interaction statistics help
SummaryThis paper proposes Unlabeled Feature Extension (UFE), 12 local statistics computed from each Ethereum account's incoming and outgoing call and transfer edges without using class labels. UFE is added to the 14 source features of HEIG-Phish and tested with six model families under matched controlled comparisons. Across 20 stratified partitions, UFE raised test Macro-F1 by 1.0 to 9.1 percentage points. An ablation associates most of the gain with the incoming transfer-edge sum attribute.