Peer groups that actually catch anomalies
Compare a merchant to the wrong 'normal' and the anomaly disappears. Measured: 91% of cases caught with peers built on identity, 24% with peers built on behavior.
Work
Case studies
Open-data projects, each with a public write-up, reproducible analysis, and interactive results.
Demand forecasting & reorder simulation for an online retailer
SKU-level demand forecasts feeding a reorder/safety-stock simulator — the £ cost of forecast error made visible.
Professional experience
Client and employer work is covered by NDAs — described at a high level only, with each problem class linked to an open case study that demonstrates it.
Transaction-monitoring anomaly detection (banking)NDA
Anti-money-laundering anomaly detection for Italian banks: peer-group clustering, isolation forests, and behavioral feature engineering on transactional data.