Estimating Parcel Delivery Day via Quantile Regression

The problem of delivery time estimation consists in accurately predicting how long it will take for a parcel to be delivered to a customer. These predictions have traditionally been computed by analyzing large collections of carrier, parcel and customer data (e.g., shipping method, carrier performance along a given route, parcel size/weight, buyer/seller address, GPS location, among others). In this paper, we discuss a quantile regression framework to predict the time interval where a delivery is most likely to happen. We do this under constraints of data minimization in agreement with privacy regulations such as the EU-GDPR. We compare aggregations of historical data using Eurostat’s Nomenclature of Territorial Units for Statistics (NUTS) with our proposed method based on quantile regression with LightGBM and two simple Bayesian regression models. Our framework achieves competitive performance in delivery time estimation while safeguarding user privacy.