The rapid growth of digital finance and e-commerce has led to an unprecedented surge in online transactions, making financial systems increasingly vulnerable to sophisticated and evolving fraudulent activities. Despite the deployment of conventional models in fraud detection systems, most existing approaches lack the adaptability to detect dynamic fraud patterns, suffer from high false positives, and are incapable of capturing complex spatial–temporal relationships in transaction data. To address these gaps, this work presents a cloud-enabled hybrid Transformer-CNN for online payment fraud detection using optimal pattern analysis. Transformer-based models—TabTransformer, TransTab, BERT, and RoBERTa—to extract temporal and contextual features from structured online transaction data. To address redundancy and irrelevance in high-dimensional features, an enhanced artificial flora optimization (EAFO) algorithm is used for optimal feature selection, thereby enhancing the discriminative power of the model. The refined feature patterns are then classified using a spatial deep convolutional neural network (SD-CNN), which effectively captures spatial correlations within the optimized feature space to enable accurate fraud detection. Fraud detection outcomes are sent to a cloud system, where each transaction is securely verified using blockchain to ensure transparency, traceability, and tamper resistance.The proposed model is evaluated on three widely used real-world datasets: PaySim, IEEE-CIS, and the European cardholder dataset. The proposed hybrid RoBERTa + EAFO + SD-CNN model outperforms existing approaches with a performance gain ranging from 8.2 to 15.6% in key metrics such as accuracy, AUC, precision, and recall across multiple real-world datasets. Furthermore, the integration of a blockchain-enabled cloud layer ensures secure and tamper-proof transaction logging, enhancing data integrity by 22.4%, traceability by 19.7%, and transaction auditability by 25.3%. These enhancements make the solution not only highly accurate but also robust, secure, and scalable for real-time deployment in modern cloud-based digital banking environments.
Fnu et al. (Fri,) studied this question.