Virginia Tech graduate students team up with D.C. transit to help enhance customer service
Last fall, the Washington Metropolitan Area Transit Authority (WMATA) struck a partnership with Virginia Tech’s graduate program in urban computing for help in predicting its system’s on-time performance (OTP).
The resulting study, by a team of students enrolled in Introduction to Urban Computing, a computer science course in the UrbComp certificate program administered by the Discovery Analytics Center, is one of the first steps in connecting WMATA’s Rush Hour Promise -- initiated in January 2018 to provide a refund to any customer delayed by 15 minutes or more during rush hour -- to underlying service disruptions, according to Jordan Holt, senior performance analyst at WMATA.
Holt said that the research could also help WMATA better target customer communications about delays and tie late customers to particular disruptions.
“Right now, we can look at individual pieces of information in our incident and customer databases and make very good estimates, but we do not have a systematic way of marrying our incident database with our databases that track customer and train movement to calculate the system-wide impact of service disruptions,” Holt said.
The students’ approach developed predictive models that could enable forecasting the experience of riders currently in network and riders who have not yet entered the system. Prior to this, WMATA’s analysis was limited to retrospective studies of OTP.
“We look forward to continuing to work with the UrbComp program to see how we can apply the students' approach to improving our operations,” said Holt.
WMATA, operating 1,126 railcars on six different routes over 234 miles of track to support more than 500,000 passengers each day, defines OTP as the percentage of passengers whose trip time is less or equal to WMATA’s Travel Time Standards. As an example, according to these standards, it should take no longer than 15 minutes to travel from Ballston to the West Falls Church Metro station during morning peak time. Customers who experience a trip longer than this time are said to be late and decrease OTP for the network.
For the study, WMATA provided the team with anonymized tap-in/tap-out passenger information collected from turnstiles gating entry to and exit from the stations or the Metrorail network, as well as travel time standards for estimating OTP trip times. A year’s worth of data totals over 150 million records.
The team also gained access to data about train interactions with station platforms. This data spanned nearly a two-year period from Jan. 4, 2015, to Dec. 31, 2016, and included 36,624,940 records, each accounting for a single arrival/departure of a train at a network platform.
Using both sets of data, the students focused their attention on a subset of months to avoid major long-term maintenance efforts and holiday scheduling, and developed deep learning algorithms to predict OTP.
The student researchers on the WMATA project were: Sandeep Kumar Bijinemula, of Washington, D.C., and Bryse Flowers, of Botetourt County, Virginia, both master’s degree students in electrical and computer engineering; Farnaz Khaghani, of Iran, a Ph.D. student in civil and environmental engineering; and Randy Soper, of Manassas, Virginia, a master’s student in computer science at the Discovery Analytics Center.
“One of the best aspects of working with WMATA, and the UrbComp initiative at Virginia Tech in general, is its interdisciplinary nature,” said Flowers. “Working hand-in-hand with Jordan Holt at WMATA, we were able to understand what constitutes a ‘good’ trip on the Metrorail system and how that impacts business decisions made by WMATA at a high level and in real time by its operations control center.”
Flowers said that the students were then able to apply their academic training in data analytics and machine learning to a real-world dataset that illustrated the positive and immediate tangible impact these skills can have on problems that may at first seem to be out-of-scope for their domain.
“The students’ work was of such high quality that they have been able to submit the study as an academic paper,” said Holt. “I was very impressed by their ability to quickly become familiar with WMATA’s data so that they could understand how to best apply the deep learning methodology to help WMATA determine how delays propagate through the rail network.”
“This collaboration with WMATA is indicative of the hands-on data science experience we aim to impart to our UrbComp students,” said Naren Ramakrishnan, the Thomas L. Phillips Professor of Engineering, director of the Discovery Analytics Center, director of the National Science Foundation-sponsored UrbComp program, and instructor for Introduction to Urban Computing.
The research project and interactions with WMATA were overseen by Brian Mayer, project manager and research scientist at the Discovery Analytics Center. “It has proven to be a win-win situation for both the students and our partner,” Mayer added.
From a student perspective, Khaghani agrees. “The project-based approach of this course made us learn major concepts of urban computing through hands-on experience with a real-world project, which I consider to be one of the best experience of my graduate studies,” she said.
Written by Barbara L. Micale