AI / MLOctober 2026
Delivery Risk & Demand Models
Models that predict late-delivery risk and stop handling time, forecast weekly demand and plan vehicle allocation for peak days, integrated into the Waypoint Relay delivery system. On time-based validation they reached about 0.98 AUC for late-delivery risk and about 3.6% error on weekly demand.
PythonLightGBMCatBoostScikit-learnSHAPStreamlit
About the project
Delivery Risk & Demand Models is a set of machine learning models that predict how long each delivery stop will take and how likely it is to run late, forecast weekly demand, and plan the best vehicle allocation for peak days. They were then integrated into the Waypoint Relay delivery planning system.
Key features
- Late-Delivery RiskLightGBM and CatBoost ensemble with an AUC of about 0.98 on time-based validation
- Service Time PredictionEstimates handling minutes for every planned stop
- Demand ForecastBottom-up weekly forecast with an error of about 3.6%
- Vehicle Allocation OptimizerExact optimization that respects every delivery rule on peak days
- Demo DashboardsStreamlit apps for dispatchers to explore risk and forecasts