Pre-Transaction Fraud Risk Prediction in DeFi Using Explainable AI

Abstract

Decentralized Finance (DeFi) has changed the financial ecosystem but is extremely vulnerable to complex schemes such as rug pulls, honeypots, and flash loan attacks, which cause massive losses for platforms and users. This study responds to the critical necessity of early and actionable warning for fraud by suggesting a real-time system predicting and justifying the risk level of freshly launched DeFi tokens prior to their engagement with the users. Drawing on a hybrid methodology, integrating smart contract code analysis, on-chain behavioral data, and social metrics, the system utilizes cutting, edge machine learning models, including Graph Neural Networks (GNNs) and ensemble methods (XGBoost, FT-Transformer), to provide sophisticated risk scoring. For user and regulatory trust assurance, explainable AI methods such as SHAP and LIME are utilized to clearly identify important risk drivers, including unsafe contract functions, wallet concentration, and liquidity lock patterns. The solution provides stage-aware and cross-chain surveillance, coupling functionality like anomaly detection, federated model training, and governance analysis. With large-scale literature synthesis and empirical validation, this framework shows that proactive, pre-transaction fraud identification and open risk valuation are possible, and it achieves a scalable defense for DeFi players and infrastructure.