Bank Predictive Analytics

Churn, credit risk and remittance forecasting for a synthetic Saudi retail bank. Models are trained and validated out of time by the pipeline; this page shows its gold-layer output for a 20,000-customer sample.

Source code on GitHub · all data is synthetic

Balance at risk by segment (SAR)

Expected loss by risk grade (SAR)

Total daily remittance — last 180 days and next 90 days (SAR)

High-risk customers by primary signal

Average churn probability by risk decile

Colour = risk band: red High (decile 10), amber Medium (8–9), green Low.

Retention worklist — highest churn probability first (top 50)

EAD by risk grade (SAR)

Average 12-month PD by DBR band

Watchlist — largest expected loss first (top 50)

Daily remittance: actual, backtest and forecast with P10–P90 band (SAR)

Peaks: Saudi payday window (~27th), Saturdays and the days before Eid. Last 180 days of history shown.

90-day forecast by corridor (SAR)

Permutation importance

Drop in AUC when the feature is shuffled (out-of-time sample).

All metrics