back
Credit Card Fraud Detection
service : ML engineering
timeline : March 2024
role : ML engineer
A real-time fraud detection system designed to identify fraudulent credit card transactions as they occur, minimizing financial losses for institutions.
problem
High latency in existing fraud detection systems led to significant financial losses. By the time fraud was detected, transactions had already been processed and funds transferred.
approach
Built a real-time inference pipeline using Apache Kafka for stream processing and Isolation Forests for anomaly detection. Optimized model serialization using ONNX Runtime to achieve sub-50ms latency per prediction.
outcome
Achieved 98% detection rate for fraudulent transactions in real-time. Reduced average detection time from minutes to under 50 milliseconds.