Passive optical network optimization and analysis using toward trustworthy fintech: a blockchain-BiLSTM approach for accurate, auditable fraud detection

Abstract FinTech fraud has become a pressing challenge as financial services increasingly migrate online. While digital platforms have transformed banking, payments, lending, and insurance, they have also expanded the attack surface for fraudulent activities. Industry studies estimate that organizations lose about 5 % of annual revenue to fraud, with global losses projected at US$5–5.5 trillion annually. The surge in digital payments further exacerbates this problem; for instance, the U.S. Automated Clearing House (ACH) processed over 33 billion transactions in 2024, exceeding manual auditing capacity. Insurance fraud, particularly in healthcare, also remains substantial, consuming 3–10 % of expenditures. To address these challenges, this paper proposes an integrated fraud-detection framework that combines blockchain technology with bidirectional long short-term memory (BiLSTM) neural networks. Blockchain offers transparent, tamper-proof, and auditable ledgers for secure data sharing, while BiLSTM models capture sequential dependencies in transactions by learning from both past and future contexts. This integration overcomes the limitations of rule-based, isolated AI, and purely cryptographic methods that often fail against sophisticated schemes. Experimental evaluation on simulated credit card and lending datasets demonstrates significant improvements in fraud detection accuracy, auditability, and false-positive reduction. The proposed blockchain–BiLSTM framework enhances trust, compliance, scalability, and security, providing a robust foundation for next-generation FinTech fraud prevention.