Generative AI-based Framework for Fraud Detection and Prevention in Online Payment Systems

The escalating menace of financial fraud has emerged as a significant issue in the contemporary digitally interconnected landscape, particularly with the proliferation of e-commerce and online payment mechanisms. Fraudulent activities, especially in credit and payment card transactions, have increased significantly, leading both public and private sectors to invest substantially in research and development aimed at fraud detection and prevention in online transactions. This paper presents a comprehensive strategy that initiates with a rigorous data preprocessing pipeline to manage missing values, eliminate outliers, and normalise the dataset through undersampling approaches. Fraud indicators are derived using dimensionality reduction, and the Improved Red Piranha Optimisation (IRPO) algorithm is utilised to optimise feature selection. A hybrid spatial-temporal feature space is created by amalgamating datasets from several sources, including sensors and services. Multiple deep learning algorithms, such as LSTM, GRU, and a hybrid LSTGR model, are subsequently employed on time-series geographic data for the purpose of fraud detection. The hybrid LSTGR model exhibited exceptional performance, with an accuracy of 99.08%. The results underscore the efficacy of integrating advanced preprocessing, optimisation, and deep learning methodologies for precise and efficient fraud detection and prevention in online financial transactions.