Public Transport Commuting Analytics: A Longitudinal Study Based on GPS Tracking and Unsupervised Learning

Abstract

This paper uses unlabelled GPS tracking data collected by a smartphone application, enriched by fusion with automatic vehicle location (AVL) data, to study commuting trips from home to work and vice versa. Such commuting trips play a significant part in public transport (PT), and in transport planning in general. This work investigates patterns of mobility, based on multiple thousands of recorded trips over a set of users in a longitudinal study by, first, determining unsupervised clustering algorithms to impute work and home locations, then analysing relevant characteristics, such as departure times, mode/line choice and trip duration. Finally, a heuristics algorithm is proposed to analyse the extent and frequency of similar trips. The results quantify amount and limits of the regularity of individual commuting behaviour in terms of repeatable travel choices. Commuters are quite consistent in their choices of departure times and lines used, even though differences are found among the two directions of the commuting trips, with work–home trips having a greater average duration and, in many cases, different choices of lines.