← Back to projects
Biznesni Rivojlantirish Banki
Production
Real-Time Fraud Detection System
Biznesni Rivojlantirish Banki⚡Fraud and anomalous transactions had to be identified in real time - any delay means the fraudulent transfer has already settled and recovery is far harder.
⚠️Legacy checks were rule-based and static - they missed novel fraud patterns and generated high false-positive noise that buried real signal.
⚙️Built an anomaly-detection pipeline scoring each transaction against learned customer behaviour - engineered features from transaction sequences, amounts, timing, and account profiles.
🛡️Addressed extreme class imbalance (fraud is a tiny fraction of volume) with resampling and threshold tuning, optimising catch-rate without flooding analysts with false alerts.
🚀Deployed as a real-time flagging service - suspicious operations are surfaced to the security team automatically, reducing manual transaction review and tightening the bank's defences.
↳ Real-time detection of fraudulent and anomalous transactions in production - suspicious operations flagged automatically, strengthening bank security
Anomaly DetectionReal-Time ScoringScikit-learnFeature EngineeringImbalanced DataPython