Over the last ten years, transit riders have learned to expect accurate real-time data. Research has shown that real-time information about when the next bus will arrive can reduce passengers’ perceived waiting time and even reduce actual waiting time, allowing passengers to time their arrival at the stop to when the bus will actually arrive. In New York City, where Cambridge Systematics (CS) pioneered real-time information with the Metropolitan Transit Authority (MTA), studies showed that bus ridership increased after MTA Bus Time® launched. On the other hand, when your phone says the bus should arrive in four minutes and it doesn’t show up for seven minutes, you may feel that your transit agency is letting you down.

We are excited for the chance to work with Minneapolis’ Metro Transit for an ongoing pilot to improve their real-time predictions. In an article from the Star Tribune, Metro Transit highlighted that an increasing number of its riders rely on its NexTrip information, particularly in harsh weather. Based on customer feedback, as well as internal measures, Metro Transit identified the need for better predictions. They structured a pilot program to evaluate competing approaches and pick the best technology for the agency and its riders.

In partnering with transit agencies across the U.S. to deliver real-time data, our team at CS knows that the challenges can be surprisingly complex. The technology must deal with the details of local transit service, streets, and traffic. All the while, the solution needs to be stable enough to serve tens of thousands of simultaneous users on both the agency’s apps and third party apps without lagging or dropping. Addressing these issues requires a close and ongoing partnership between technology experts and transit operators to tune the software for local conditions.

At CS, we pride ourselves on our use of open source technologies because it allows us participate in a global community of developers working on the same challenges. To tackle the problem for Metro Transit, we’re building a solution based on TheTransitClock, an open source engine that uses real-time positions to generate arrival predictions. We’re teaming up with experts Sean Óg Crudden and Simon Berrebi, who are independent leaders in maintaining and enhancing TheTransitClock. Together, the team will deliver a customized real-time prediction algorithm that combines historical and real-time information with Metro Transit-specific operating rules and practices to produce the most accurate results. This prediction method uses an adaptive algorithm, so the predictions will improve over time and react to unexpected events.

The prediction algorithm employs a Kalman Filter that draws on data from the past weeks, days and hours to handle travel time and dwell time (when the bus is stopped). The Kalman Filter is an adaptive algorithm, capable of recognizing unusual conditions (such as traffic congestion, weather events, etc.). When these events occur, the algorithm incorporates the travel times of downstream vehicles to create predictions that are more accurate than traditional methods.

Because the prediction algorithm considers travel times and dwell times separately, it can predict situations that lead to bus bunching—when several buses show up together after a long wait. When a bus is preceded by a large gap, perhaps because it is running late, the algorithm anticipates that dwell times will be longer than usual at the next stop. This helps to manage passenger expectations and gives supervisors enough time to intervene.

CS will be running TheTransitClock pilot through October and working iteratively with Metro Transit to assess accuracy and continuously improve performance. If all goes well, riders will have better predictions starting this winter and won’t have to wait longer for the bus in the snow than they expect —and in Minnesota winters, that’s no small thing.