Machine Learning Models for Fraud Detection in Credit Cards in the Banking Sector

Credit card fraud pose a growing threat to financial institutions as digital transactions become rapidly complicated and wider. Traditional rules-based systems often fail to detect fraud patterns that mimic valid behavior. In response, machine learning (ML) and deep learning (DL) models have emerged as a adaptive tool capable of identifying discrepancies in real time. Ayub (2023) displayed that the LSTM models perform better than the traditional classifier, receiving the AUC-RC of 0.987 and the F1-Score of 0.909. Alatawi (2025) showed that integrating the IOT-promoted features into a contingent of artists improves the accuracy to detect to a large extent. Tang (2024) introduced the structured data transformer (SDT) with federed learning, balanced interpretation and privacy. Babdullah (2024) proposed a blockchain-acquired federed framework using random forests, enabling safe multi-bank cooperation. Alazharani (2024) strengthened the value of decentralized learning and blockchain for privacy-protection deployment. This review synthesize these contributions, identifies gaps in hybrid architecture, and proposes future directions for scalable, clear and institutionally obedient fraud detection systems.