loading page

Order acceptance choice modeling of crowd-sourced delivery services: a systematic comparative study
  • Shixuan Hou
Shixuan Hou
Concordia University

Corresponding Author:[email protected]

Author Profile

Abstract

The efficiency of crowd-sourced delivery services (CDS) like UberEats and AmazonFlex highly depends on the decisions of individual shippers. Operating as freelancers, these shippers have the freedom to accept or decline orders from the CDS platform. Their decisions not only affect their earnings and the waiting times for orders but also influence the platforms’ overall revenue and reputation. Understanding the factors that shape these decisions is thus crucial. In our study, we gather data from CDS shippers in Shanghai, China, using stated preference surveys. We then design a discrete choice model to predict shippers’ behaviors and compare its accuracy, computational efficiency, and interpretability with five commonly used machine learning methods. Our analyses reveal that the Extreme Gradient Boosting (XGB) model and Random Forests (RFs) model outperform other models in prediction accuracy, achieving f1 scores of 69.3% and 65% respectively. Notably, our per- mutation importance analysis indicate that the shipper’s age, income, and the compensation awarded per order are the most influential determinants in their decision to accept or decline orders.