Preto, A., Schneider, A., and Bierlaire, M. (2026)

Heterogeneity in Mobility Behavior: Latent Classes from Long-Term Tracking Data

26th Swiss Transport Research Conference, Ascona, Switzerland

Understanding mobility behavior requires going beyond mode choice and trip counts to capture how individuals structure their daily and weekly activities in time and space. Routine, flexibility, and variability in travel patterns influence public transport demand and peak congestion. We address this challenge by exploiting long-term passive tracking data from the Continuous Mobility Panel dataset, comprising approximately 2,000 individuals observed continuously between 2023 and 2025. However, tracking data at daily resolution are noisy: without careful modeling, we risk estimating noise rather than structure. To extract meaningful patterns, we use a hierarchical latent class model with individual random effects that explicitly separates between-class structure, persistent individual heterogeneity, and day-to-day variability. The latent classes reveal distinct and interpretable mobility patterns, including routine schedule-constrained commuters, high-mobility users and hybrid workers combining regular anchors with flexible schedules. Rather than relying on a priori segmentation, class membership is inferred from observed mobility behavior. This framework allows the introduction of structural equations linking class membership probabilities to socio-economic characteristics. The results provide insights for public transport planning. Class-specific demand profiles identify which user groups contribute to peak loads and which exhibit greater temporal or spatial flexibility, and are therefore more likely to adjust departure times or routes in response to crowding. Overall, the proposed latent class approach offers a behavioral alternative to traditional marketing personas, enabling a deeper understanding of mobility routines and their implications for public transport system design and operation.