MIT Transit Lab

Origin-destination matrices provide vital information for service planning, operations planning, and performance measurement of public transportation systems. In recent years, methodological advances have been made in the estimation of origin-destination matrices from disaggregate fare transaction and vehicle location data. Unlike manual origin-destination surveys, these methods provide nearly complete spatial and temporal coverage at minimal marginal cost. Early models inferred destinations on the basis of the proximity of possible destinations to the next origin and disregarded the effect of waiting time, in-vehicle time, and the number of transfers on path choice. The research reported here formulated a dynamic programming model that inferred destinations of public transportation trips on the basis of a generalized disutility minimization objective. The model inferred paths and transfers on multileg journeys and worked on systems that served a mix of gated stations and ungated stops. The model is being used to infer destinations of public transportation trips in Boston, Massachusetts, and is producing better results than could be obtained with earlier models.

For public transportation agencies, the ability to provide personalized and dynamic passenger information is crucial for improving the efficiency of demand management and enhancing customer experience. This requires understanding and especially predicting individual travel behavior in the public transportation system, which is challenging because of the heterogeneity among passengers and the variability of their behaviors. This paper presents, to the best of the authors' knowledge, the first attempt to predict individual spatiotemporal behavior of public transportation passengers using smartcard data. In this study, each trip is coded as a combination of trip start time, an entry station and an exit station. A passenger's daily mobility is represented as a chain of travel decisions. The authors propose a new modeling framework, inspired by Bayesian n-gram models used in natural language processing, to estimate the probability distribution of the next decision in the sequence. Empirical analysis using Oyster card data from London shows promising results. It is found that the exact time of travel is most challenging to predict, but the difference between the predicted time and the true value is usually small. Model performance varies greatly across individuals for the prediction of entry and particularly exit stations. Overall, the proposed model shows significant improvement over the regular n-gram models, or Markov chain-based models in general. The improvement is even larger for weekend trips when travel behavior is flexible, irregular, and considerably less predictable.

Although automatically collected human travel records can accurately capture the time and location of humanmovements, they do not directly explain the hidden semantic structures behind the data, e.g., activity types. This work proposes a probabilistic topic model, adapted from Latent Dirichlet Allocation (LDA), to discover representative and interpretable activity categorization from individual-level spatiotemporal data in an unsupervised manner. Specifically, the activity-travel episodes of an individual user are treated as words in a document, and each topic is a distribution
over space and time that corresponds to certain type of activity. The model accounts for a mixture of discrete and continuous attributes---the location, start time of day, start day of week, and duration of each activity episode. The proposed methodology is demonstrated using pseudonymized transit smart card data from London, U.K. The results show that the model can successfully distinguish the three most basic types of activities---home, work, and other. As the specified number of activity categories increases, more specific subpatterns for home and work emerge, and both the goodness of fit and predictive performance for travel behavior improve. This work makes it possible to enrich human mobility data with representative and interpretable activity patterns without relying on predefined activity categories or heuristic rules.