Final YearAdvancedData 3-4 months
Credit-Card Fraud Detection System
A real-time fraud detector that scores transactions using ML on highly imbalanced data and flags suspicious activity through a monitoring dashboard.
Pythonscikit-learn / LightGBMKafkaFastAPIReact
Find GitHub repos & source code →How to build it — step by step
- 1Imbalance: Engineer features and handle extreme class imbalance with sampling/cost-sensitive learning.
- 2Modelling: Train and evaluate with precision-recall and PR-AUC rather than raw accuracy.
- 3Streaming: Score transactions in real time over a Kafka stream with low latency.
- 4Monitoring: Surface flagged transactions, alert thresholds, and feedback for retraining.
Key features to implement
- ✓Imbalanced-data handling
- ✓Real-time scoring
- ✓PR-AUC evaluation
- ✓Alerting dashboard
- ✓Analyst feedback loop
💡 Unique twist to stand out
Add concept-drift detection that monitors score distributions and triggers retraining when fraud patterns shift.
🎓 What you'll learn
Imbalanced learning, streaming inference, proper evaluation metrics, and MLOps monitoring.