FERDINAND, SOUW TIMOTIUS ALVIN (2026) HYBRID FRAUD DETECTION ON ETHEREUM BLOCKCHAIN USING ML AND SNA. S1 thesis, UNIVERSITAS KATOLIK SOEGIJAPRANATA SEMARANG.

Preview	Text 22.K1.0001 - SOUW TIMOTIUS ALVIN FERDINAND-COVER_a.pdf Download (684kB) | Preview
	Text 22.K1.0001 - SOUW TIMOTIUS ALVIN FERDINAND-ISI_a.pdf Restricted to Registered users only Download (1MB)
Preview	Text 22.K1.0001 - SOUW TIMOTIUS ALVIN FERDINAND-DAPUSI_a.pdf Download (726kB) | Preview
	Text 22.K1.0001 - SOUW TIMOTIUS ALVIN FERDINAND-LAMP_a.pdf Restricted to Registered users only Download (1MB)

Abstract

As Ethereum grows rapidly, blockchain fraud like phishing and illegal fund transfers has surged. Traditional detection relies heavily on transaction statistics, ignoring relational interaction patterns between suspicious wallets. This study proposes a hybrid framework integrating Machine Learning (ML) and Social Network Analysis (SNA) to enhance fraud detection within the Ethereum ecosystem. Transaction data were transformed into a directed graph where wallet addresses represent nodes and transactions represent directed edges. Two feature sets were extracted: transaction-based behavioral features (frequency, value, diversity, burst behavior) and graph-based structural features (Degree, Betweenness, PageRank, Clustering Coefficient). These combined features trained several models, including Random Forest, XGBoost, and Isolation Forest. Experimental results demonstrated that integrating SNA features significantly improves performance over transaction-only approaches. The proposed Hybrid Random Forest + SNA model achieved the best overall performance with an accuracy of 0.9987, F1-score of 0.9956, and ROC-AUC of 0.99998, outperforming the baseline transactiononly model. Graph visualization revealed that suspicious wallets frequently occupy highly connected, influential, and bridge-like positions within the transaction network. These findings indicate that combining ML and SNA provides more robust, interpretable, and context-aware capabilities for identifying structurally suspicious blockchain behavior..

Actions (login required)

View Item