Moving from classical towards machine learning stances for bus passengers’ alighting estimation: a comparison of state-of-the-art approaches in the city of Lisbon

Passengers alighting estimation is a critical task in public transport (PT) management, especially for entry-only Automatic Fare Collection (AFC) transport systems where passengers alighting are not recorded. Effective estimation methods are necessary for trip analysis and route planning, offering valuable insights into passengers’ mobility patterns and, subsequently, improving the quality of service. However, the stochastic nature of passenger behaviour challenges the degree of successful alighting estimates. A classic approach to infer the alighting stops of passengers is the use of trip-chaining principles. Since these principles are dispersed across the literature in the field, their comprehensive review is pivotal to establish the best practice for alighting estimation. Still, trip-chaining approaches are unable to infer the alighting of non-commuting passengers. This paper addresses these two research gaps by: i) providing a critical overview of the existing principles and methods for alighting estimation; ii) proposing an approach to improve alighting estimation that consistently integrates the most effective state-of-the-art principles on trip-chaining; and iii) further introducing a frequent pattern mining and density-based clustering solutions to support alighting estimation for non-commuting passengers. Considering the public bus transport in Lisbon city as the guiding case study, the achieved estimation rate by the proposed assembled model is 92%. Moreover, the density-based clustering solution is found to improve the estimation of 11pp against classic trip-chaining principles. Furthermore, the proposed model and acquired results yield actionable value to enhance PT operations and services, ultimately leading to improved bus routing and quality of service.