A Blockchain-Anchored Fraud Detection System to Prevent Insider Tampering of Machine Learning Audit Logs

Traditional enterprise fraud detection systems often fail not because of poor algorithms, but because privileged operators can manipulate audit logs or bypass internal workflows. This research addresses the resulting trust gap by anchoring machine learning predictions and decision-making processes to an immutable blockchain ledger. By treating the ledger as an active enforcement mechanism rather than passive storage, the system ensures that every transaction and explanation is cryptographically verifiable and resistant to retroactive modification. The proposed architecture enforces approval processes through smart contracts, providing a transparent audit trail that meets regulatory requirements such as GDPR Article 22. The detection module maintains high performance, achieving an F1-score of 0.895 and a PR-AUC of 0.974. These results suggest that integrating blockchain-based verification does not necessarily compromise the accuracy of the underlying machine learning models compared to legacy centralized systems. Operational evaluations demonstrate the system's readiness for enterprise-scale workloads, featuring inference latencies under 25 milliseconds. Deployment on Layer-2 networks enables cost-effective execution at less than $0.01 per transaction, supporting workloads of over 10,000 monthly payments. Engineers considering this approach should evaluate their specific threat models to determine if the added security of decentralized anchoring outweighs the overhead of smart contract integration.