AI-Driven UCCGAN Framework for Detecting and Preventing Fake Check Scams

Abstract

Check fraud has become a major menace within current banking systems and has caused huge financial losses and mistrust of financial dealings. Conventional fraud detection techniques tend to be either rule-based or supervised learning techniques, are constrained by the presence of labelled data, and are not flexible to changing fraud patterns. To overcome these issues, this study introduces a smart architecture, AI-UCCGAN-DST-FCS, that combines the advantages of blockchain-based authentication with unsupervised deep learning to effectively detect fraud. First, to verify the integrity of the data and establish trust among the involved banks, cheque transactions are verified using a proof-of-green (POG) consensus mechanism inside a blockchain network. An adapted Unsupervised Cycle-Consistent Generative Adversarial Network (UCCGAN) is then used to learn complicated patterns in transactions and identify anomalies without any labelled data. In addition, the Secretary Bird Optimization Algorithm (SBOA) was applied to optimize the parameters of the model, improving the detection accuracy and computational efficiency. The experimental analysis of publicly available financial fraud dataset shows that the proposed approach has better performance, with an AUC of 0.984 and greatly reduced error rate and computation time in comparison with the existing methods. The findings affirm the suitability of the proposed system for detecting cheque fraud in real time and securely.