Integrating MLOps, SRE, and Chaos Engineering for Scalable and Resilient Credit Card Fraud Detection Systems

The soaring expansion of digital financial transactions creates an ever-pressing need for effective and scalable credit card fraud detection techniques. Traditional machine learning models, which perform admirably in a contained environment for development, are not able to maintain their performance and resilience under gradually changing production environments. In this work, the paper recommends an end-to-end system that builds on novel integration of Machine Learning Operations (MLOps), Site Reliability Engineering (SRE), and Chaos Engineering to develop a fraud detection pipeline that is entirely READY for production. Based on a logistic regression model trained on a simulated dataset of credit card transactions, the system achieves high accuracy while offering automated deployment, performance monitoring, and fault injection capabilities. The model's proved effectiveness having scored 99.64% accuracy, 99.75% precision, and an AUC-ROC score of 0.9975. Furthermore, stress testing substantiated the system's scalability and fault tolerance through latency and throughput benchmarks. The newfound amalgamation of operational reliability and machine-learning-based anomaly detection sets a new standard within fraud analytics for banks.