Build model to predict package delivery time

Treat this as a conditional time-to-event forecasting problem. At a prediction timestamp $$t_p$$, define the target as remaining delivery time: $$y = t_{\text{delivery}} - t_p$$ Equivalently, the product-facing ETA is: $$t_{\text{delivery}} = t_p + \hat y$$ At training time, use only completed shipments whose delivery timestamp is known and whose features are available as of $$t_p$$. 1. Prediction target and timing Define the target as remaining time until delivery, not…