AI-Driven Fraud Detection in Real-Time Payment Processing Systems: A Java-Based Approach

Abstract

Fraud schemes are becoming more sophisticated and digital payment systems are exposed so there is a need for accurate detection methods that can work in real time. We introduce an AI-based framework for detection of fraud that cooperatively utilizes a deep neural network (DNN) model and domain specific heuristics based on rules. It is a microservices-based architecture, all nice and wrapped in Java with Spring Boot powered by the asynchronous, real-time streaming of Apache Kafka. The tool is a custom Java simulation engine that creates a dataset of 1.5 million transactions with realistic features and controlled fraud injections. Extensive technical performance metrics indicate 97.1% detection accuracy, 95.4% recall, and an average transaction processing delay of 78 ms. This work covers each step from data simulation and AI model design to enterprise integration, providing a foundation for realistic deployment in high-throughput financial environments.