LAUSR

This study presents a model system for trip destination prediction with multi-day GPS data. The pre-trip destination prediction model and the during-trip destination prediction model are constructed and calibrated for trips on weekdays and weekends, respectively. Combining Markov chain and Multinomial logit model, the habit of multi-day destination choice is learned and utilized in developing the pre-trip model. By introducing support points, the during-trip model is developed with the Hidden Markov model. The estimation results indicate that this study improves the precision for destination prediction by calibrating the models for trips on weekdays and weekends separately and considering habit of multi-day destination choice. The findings suggest that travelers destination choice behavior follows not only the continuity between adjacent destinations, but also the inertia at the same time of day in consecutive days, and consecutive weeks as well. The quantitative comparison of the habit-based factors indicate that the inertia between adjacent destinations has the greatest effect on weekends’ destination choice, and that in consecutive days has the most significant impact on weekdays’ destination choice. In addition to real-time travel navigation based on the during-trip destination forecasting, the model can be applied to Advanced Traveler Information System which provides travelers with pre-trip information, such as traffic condition and commercial facilities around the destination. Through pre-trip destination prediction, study findings can also be utilized in trip distribution prediction or crowded location analysis, which presents a wide application in transportation planning and management.