🚚 Shipping & Logistics ML — live demo
Three use cases from the open-source repo, driven by the controls below. Build a shipment and score it two ways, then see how a fixed intervention budget should actually be spent. All models are trained in-process on documented synthetic generators — no real customer data, and every number here reproduces the repo's tests.
Build a shipment
Set the operational conditions on the left, then score it. The miss-risk model and the ETA model both read the same shipment, using only information known at induction time (no cheating with in-transit scans).
Spend an intervention budget
A risk score is not a decision. Given a day of 20,000 shipments and a fixed daily budget, which ones do you reroute, upgrade, or leave alone? Move the budget and compare the policies. Expected-value greedy weighs each shipment's risk against the cost of a missed delivery for that customer; top-K just flags the scariest scores.
About this demo
Part of Shipping and Logistics Use Cases, twelve self-contained, end-to-end machine-learning projects for parcel and freight operations. Each ships with a documented synthetic generator, audited cleaning, an honest operational baseline, evaluation in dollars and days, and explainability grounded by tests.
- Miss risk comes from
delivery-commit-prediction(XGBoost + SHAP). - ETA comes from
eta-regression(XGBoost quantile models, monotone-rearranged). - Budget comes from
intervention-optimization(expected-value allocation, counterfactually evaluated).
The full repo also covers volume forecasting, route optimization, dynamic pricing, predictive maintenance, address resolution, returns prediction, capacity planning, network anomaly detection, and exception triage — several validated on real public datasets (Olist, UCI AI4I, CVRPLIB). The models here are trained small for a fast demo; the repo's reported numbers come from full runs.