Design Real-Time Credit Card Fraud Detection System
Overview
This question evaluates a candidate's competency in designing real-time credit-card fraud detection systems, testing skills in machine learning model selection (supervised and unsupervised), feature engineering for online and offline contexts, handling delayed labels and class imbalance, and architecting low-latency production pipelines with retraining and monitoring. Commonly asked in the Machine Learning domain, it probes both conceptual understanding of trade-offs (asymmetric business costs, explainability, drift detection) and practical application-level architectural and operational skills for deploying, A/B testing, and maintaining low-latency fraud-detection models in production.
Real-Time Credit-Card Fraud Detection System Design
Scenario
You are designing a real-time fraud detection system for an online payments platform that processes high-volume credit-card transactions. The system must flag or block suspicious transactions with strict latency constraints while maintaining high approval rates for legitimate users.
Task
Design a credit-card fraud-detection strategy. Specifically describe:
- Data sources and labeling strategy (including delayed feedback like chargebacks)
- Feature engineering for real-time and offline contexts
- Model choices (supervised vs. unsupervised; ensemble strategy)
- Real-time architecture and latency budget
- Retraining cadence and feedback loops
- Monitoring for model and data drift, and threshold tuning
Assume business costs for false declines and fraud losses are asymmetric, labels can be delayed (e.g., chargebacks in 30–90 days), and the system must support A/B testing and human review.
Hints
- Consider supervised and unsupervised methods, latency constraints, feedback loops, and threshold tuning.
- Address cold-start, class imbalance, explainability, and fail-safe mechanisms.
Solution
Login required