Latent-Segmentation, Hazard-Based Models of Travel Time

Growing interest in performance reliability and improving data availability is motivating a shift toward probabilistic treatments of travel time across a number of intelligent transportation system applications. Hazard-based analysis supports the development of probabilistic travel time models and latent-class-style methodologies capture how the mechanisms affecting travel time are expected to differ based on congestion status. Benefiting from rich data available for metropolitan freeway travel times in the San Francisco Bay Area, this paper studies how congestion state, traffic demand, roadway variables, and weather impact travel-time performance in a probabilistic regime. Congestion state is captured as an inferred yet unobserved segmentation in the data using latent segmentation, and hazard-based models of travel times are developed for the congested and uncongested classes. The final model represents an intuitive description of the factors that probabilistically influence travel time on freeways. The predicted aggregation shows excellent agreement with the data. With opportunities for improvement in the data sources and complexity of the latent segmentation, the final model nevertheless represents a simple yet flexible solution for understanding the relationships between travel time, traffic state, and relevant behavioral, geometric, and environmental factors.