Non-recurrent events in traffic prediction

Francisco Pereira
Technical University of Denmark

It is not uncommon that traffic prediction tools and research report very high accuracy. However, the very few such tools that exist in the market seem not to be performing as well as people would like, even though their accuracy may be in fact near the announced. There is a paradox in the field: traffic prediction is not difficult most of the time (the routine conditions), but sometimes it becomes extremely hard (the non-recurrent events), which is often when it is needed! This presentation will focus on ongoing and past work from DTU, MIT and Singapore-MIT Alliance for Research and Technology (SMART) related to treatment of non-recurrent events in a traffic prediction system.

Presentation (PDF File)

Back to Workshop IV: Decision Support for Traffic