Enhancing Cheque Fraud Detection Through Transformer Models and Optimization Techniques on Structured Textual Data

The rise of fake cheque scams poses major challenges to banking, causing financial losses and eroding trust. Traditional manual and rule-based verification methods are increasingly ineffective against adaptive fraud. We propose CF-TCRMCNN-LEOA, a Cheque Fraud Detection framework combining a Regularized Bias-Aware Ensemble Kalman Filter (RBAEKF) for robust preprocessing, FA-ResNet for high-level feature extraction, Fair Proof-of-Reputation (FPoR) for secure logging, TCRMCNN for temporal and relational modeling, and LEOA-based parameter optimization. Evaluated on the Fraud Detection Transactions Dataset, CF-TCRMCNN-LEOA achieves 98.72% accuracy, 97.93% precision, and 3.18[Formula: see text]ms latency, outperforming AFR-FD-DC, BC-CF, and BC-OCF. These results demonstrate its effectiveness, reliability, and real-time suitability, integrating advanced AI and blockchain-inspired mechanisms for superior fraud detection and secure transaction auditing.