FraudShield AI: A Real-Time Financial Fraud Detection System Using Machine Learning and Scalable Web Architecture

Abstract

Financial fraud has become a critical challenge in modern digital payment ecosystems due to the exponential growth of online transactions and increasingly sophisticated attack patterns. Traditional rule-based detection systems lack adaptability and fail to detect complex, evolving fraudulent behaviors. This study presents FraudShield AI, a machine learning-driven framework for real-time financial fraud detection that integrates predictive modeling with a scalable system architecture. The proposed approach employs a Random Forest classifier trained on the PaySim dataset, combined with domain-specific feature engineering based on variations in transactional balances, to enhance anomaly-detection capability.

The system is deployed using a full-stack architecture comprising FastAPI, Node.js, React, and MongoDB, enabling real-time prediction, secure data handling, and interactive visualization. Experimental evaluation on a highly imbalanced dataset demonstrates strong performance, achieving high precision (0.9132), recall (0.9032), F1-score (0.9082), and an ROC-AUC of 0.9957, indicating excellent discriminative capability.

The findings highlight the effectiveness of ensemble learning and behavioral feature engineering in detecting rare fraudulent events. This work presents a practical, scalable fraud detection framework that bridges the gap between theoretical machine learning models and real-world financial deployment. Future work will focus on adaptive learning, real-time data streaming, and advanced deep learning models to further enhance system robustness and scalability.