Companion Proceedings of The 2019 World Wide Web Conference | 2019
Travel Time Estimation Accuracy in Developing Regions: An Empirical Case Study with Uber Data in Delhi-NCR✱
Abstract
Travel time estimates are highly useful in planning urban mobility events. This paper investigates the quality of travel time estimates in the Indian capital city of Delhi and the National Capital Region (NCR). Using Uber mobile and web applications, we collect data about 610 trips from 34 Uber users. We empirically show the unpredictability of travel time estimates for Uber cabs. We also discuss the adverse effects of such unpredictability on passengers waiting for the cabs, leading to a whopping 28.4% of the requested trips being cancelled. Our empirical observations differ significantly from the high accuracies reported in travel time estimation literature. These pessimistic results will hopefully trigger useful investigations in future on why the travel time estimates are mismatching the high accuracy levels reported in literature - (a) is it a lack of training data issue for developing countries or (b) an algorithmic shortcoming that cannot capture the (lack of) historical patterns in developing region travel times or (c) a conscious policy decision by Uber platform or Uber drivers, to mismatch the correctly predicted travel time estimates and increase cab cancellation fees? In the context of smartphone apps extensively generating and utilizing travel time information for urban commute, this paper identifies and discusses the important problem of travel time estimation inaccuracies in developing countries.