Balancing Fraud Detection and Privacy Preservation in Decentralised Finance: A Systematic Review

Decentralised finance (DeFi) has transformed financial intermediation by replacing institutional trust with programmable, composable smart contracts. The same openness that enables auditability and innovation also creates an unusually exposed environment for fraud, protocol exploits, market manipulation and cross-chain laundering, leading to losses of approximately $16.7 billion between 2011 and early 2023. This paper presents a systematic review on DeFi fraud detection, transaction monitoring and privacy preservation. The review synthesises three linked areas: fraud and attack typologies, detection and monitoring techniques, and the degree to which privacy-preserving design is incorporated into those techniques. The findings show a clear methodological shift from static analysis, rule-based heuristics and classical machine learning towards graph neural networks, temporal graph models, real-time transaction monitoring and emerging large-language-model-assisted symbolic analysis. However, the review also identifies a persistent design gap: most detection frameworks depend on unrestricted public-ledger transparency and therefore provide limited protection for legitimate users' financial privacy. Privacy-enhancing technologies, including zero-knowledge proofs and federated learning, offer credible routes towards compliance-aware detection but remain immature, fragmented and rarely evaluated in operational DeFi settings. The paper argues that deployable DeFi fraud detection requires the integration of explainable AI, compliance-by-design and privacy-by-design architecture and real-time adaptive monitoring to address both the technical and governance failures across the protocol life cycle.