Interview: production fraud detector — daily volume 10M, current FN too high.
hardAnswer
- (1) Investigate: what patterns are FNs?
- Feature gaps?
- Concept drift?
- (2) Enrich features: velocity ratios, device / IP fingerprints, graph features.
- (3) Ensemble models: existing rules + Isolation Forest + supervised XGBoost on labeled + AE on sequences.
- (4) Add HITL: analyst-labeled cases feed back weekly.
- (5) Score threshold tuning by cost-sensitive PR curve.
- (6) Monitor precision + recall + latency + drift dashboards.
- (7) Guardrail: never lower FP-rate SLA.
- (8) Iterate.
Check yourself — multiple choice
- Just retrain
- FN investigation + enriched features + ensemble (rules + iForest + XGB + AE) + HITL feedback loop + cost-PR threshold + drift monitoring + FP-rate SLA guardrail
- Random
- Only supervised
Production fraud fix: analyze + feature enrich + ensemble + HITL + threshold + monitoring.
#interview#anomaly-detection#applications
Practise Unsupervised Learning
214 interview questions in this topic.
Related questions
- Interview: 'design an unsupervised fraud detection system'.
- Interview: model trained in region A must now serve region B — approach?
- Interview: 'design anomaly detection for a factory sensor with 200 signals.'
- Interview: 'you're asked to segment 10M customers — how?'
- Anomaly detection in time series — what changes?
- How is drift detection an unsupervised problem?