Secure Digital Payment-based Anomaly Risk Classification using AI Based Recurrent Neural Network Approach for Fintech Platform

Digital payment platforms have emerged as the foundation of Fintech applications, yet they are highly susceptible to anomalies such as fraudulent transactions, identity theft, and cyberattacks. To overcome these problems, this paper presents a framework for Secure Digital Payments-based Anomaly Risk Classification utilizing an AI-powered Recurrent Neural Network (RNN) model. The sample dataset is Digital Transactions in India, a collection of various payment records. During preprocessing, the Synthetic Minority Oversampling Technique (SMOTE) is employed to address the extreme class imbalance between fraudulent and legitimate transactions. In feature extraction, Independent Component Analysis (ICA) is used to identify statistically independent features, thereby further enhancing the classification performance of the model. The suggested RNN is based on learning temporal dependencies in sequential transaction information, which enables effective anomaly detection. Compared to traditional methods, experimental evaluation exhibits better performance, characterized by high accuracy of 98.70 %, precision of 97.84%, recall of 98.32 %, and F1-score of 98.02%. This initiative enhances the safety and trustworthiness of online payment systems in real-time Fintech.