Blockchain-Based Framework for Secure Financial Transactions and Fraud Detection in FinTech Ecosystems

FinTech fraud detection systems face two compounding challenges: the adversarial adaptation of fraudsters to known detection rules, which undermines static rule-based systems over time, and the lack of a tamper-evident audit trail that complicates post-hoc investigation and regulatory compliance. This paper proposes a Blockchain-BERT framework combining a permissioned Hyperledger Fabric transaction ledger with a fine-tuned BERT transaction-pattern model, deployed across 18 FinTech institutions processing 12,000,000 transactions (9,600,000 training / 2,400,000 test, 80/20 chronological split) including 6,000 confirmed fraud cases. The framework achieves F1 = 0.884 and false-positive rate (FPR) of 1.2% against expert-labelled fraud annotations, outperforming ML-only BERT (F1 0.764, FPR 4.8%; +15.7% F1), Blockchain-only (F1 0.684, FPR 6.4%), and rule-based detection (F1 0.584, FPR 8.4%; +51.4% [(0.884-0.584)/0.584]). Smart-contract-automated settlement reduced mean settlement time from 48.4 hours to 2.4 hours (−95.0% [(48.4-2.4)/48.4]). False-positive rate fell 85.7% [(8.4-1.2)/8.4]. Network congestion reduces F1 to 0.784 (−11.3%) and cross-border regulatory mismatch to 0.724 (−18.1%), each with a deployed mitigation.

Hali tarjima qilinmagan