ANOMALY DETECTION IN ETHEREUM TRANSACTIONS USING GRAPH NEURAL NETWORKS AND EXPLAINABLE AI
Keywords:
Ethereum, Blockchain, Blockchain network, Decentralized trading, Transaction data, Graph Neural Networks (GNNs), Explainable AI (XAI), Machine learning, SHAP, GNNExplainer, Model interpretability, Anomaly detection.Abstract
Ethereum and similar blockchain platforms help with decentralized trading, nevertheless their activity is often disrupted by strange and dishonest actions. It is hard for rule-based or classical machine learning approaches to capture how different parts of a blockchain network interact with each other. The proposed method makes use of Graph Neural Networks (GNNs) to detect anomalies in the transactions on the Ethereum network. In this system, the directed graph uses addresses as nodes and transactions form edges with the attributes of value, timestamp and which type of token is involved. We use Explainable AI (XAI) approaches called SHAP and GNNExplainer to explain and understand the model’s choices. Working on real Ethereum transaction data reveals that the system finds anomalies very accurately and also adds to the understandability of those findings, supporting fraud prevention, compliance with regulations and clear analytics on blockchain records.
