An adaptive neuro-fuzzy blockchain–AI framework for secure and intelligent FinTech transactions

Abstract

Abstract Financial systems increasingly rely on digital and distributed infrastructures, exposing FinTech services to sophisticated and rapidly evolving cyber threats. Conventional rule-based security mechanisms and static machine-learning models may struggle to detect complex fraud patterns while meeting the integrity, trust, and real-time processing requirements of financial transactions. This study proposes an Adaptive Neuro-Fuzzy Blockchain–AI framework (ANFB-AI) for intelligent and integrity-aware fraud detection in FinTech environments. The framework combines a permissioned blockchain with machine-learning and adaptive neuro-fuzzy components to address transaction integrity, behavioural uncertainty, and evolving fraud patterns. ANFB-AI employs a dual-risk assessment mechanism that integrates machine-learning-based fraud probability with neuro-fuzzy risk inference. Transaction-specific blockchain indicators—including hash validity, digital-signature validity, validator approval, and smart-contract compliance—dynamically modify the fused behavioural-risk score and influence the final accept, monitor, or reject decision. This integrity-aware coupling distinguishes the proposed framework from approaches in which blockchain serves only as a transaction-storage layer. A mathematical formulation defines blockchain-integrity assessment, adaptive risk fusion, threat classification, and final decision-making. The framework employs Proof-of-Authority consensus and was evaluated within a five-node permissioned consortium under normal-, moderate-fraud-, and high-fraud scenarios. The experimental evaluation examined fraud-detection effectiveness and transaction-processing behaviour across repeated runs, with results summarized using the mean, standard deviation, and 95% confidence interval. The findings demonstrate the functional feasibility of ANFB-AI and show that blockchain-integrity feedback contributes to risk-sensitive fraud detection under the controlled proof-of-concept conditions. Results reported in related studies are considered only as contextual benchmarks because differences in datasets and experimental protocols prevent direct statistical comparison. Further evaluation using real institutional data, larger validator networks, and production-representative workloads is required to establish the framework’s scalability and broader applicability.