Improving ride-hailing order allocation via a ranked-first policy : Case Yassir

Abstract

Ride-hailing platforms typically employ the nearest-first matching policy that prioritizes proximity while disregarding driver acceptance behavior, leading to inefficient assignments. This work proposes and evaluates a Ranked-First Policy that integrates acceptance prediction into the dispatching process, using Yassir, Algeria’s leading ride-hailing platform, as a case study. An empirical analysis of 312,216 dispatch records from Oran, Algeria, revealed systematic patterns in driver acceptance behavior influenced by economic, temporal, spatial, and experiential factors. A comprehensive feature set was engineered to capture these behavioral signals, and an XGBoost model achieved an AUC of 0.785 with 79.2% Hit@1 accuracy, correctly identifying the accepting driver as the top-ranked candidate in most cases. A counterfactual simulation against Yassir’s current ETA-based policy demonstrated substantial operational improvements: first-offer success rate nearly doubled from 43.46% to 79.2%, and average time-to assignment decreased by 72%, from 21.18 to 5.79 seconds. These result confirm that acceptance-aware matching significantly enhances efficiency by reducing rider wait times and optimizing driver allocation.