2/26/2019 Celeste Arbogast
Written by Celeste Arbogast
Above: Assistant professor Hadi Meidani and his Ph.D. advisees Mohammad Nabian and Negin Alemazkoor placed second in an international competition to design a train-delay prediction algorithm.
By Celeste Arbogast
A CEE team won second place in an international competition to develop a train delay prediction algorithm. Ph.D. candidates Negin Alemazkoor and Mohammad Nabian, and their adviser, CEE assistant professor Hadi Meidani, placed second out of 40 teams with their submission, “Predicting Near-Term Train Schedule Performance and Delay Using Bi-Level Random Forests.”
The competition is presented yearly by the Railway Applications Section of the Institute for Operations Research and the Management Sciences (INFORMS) at its annual meeting. More than 40 international teams from academia and industry competed in the 2018 contest, the goal of which was to predict delays in the Netherlands Railway passenger rail network. Based on their initial submissions, the top three teams were chosen to present their work at the INFORMS annual meeting Nov. 4, 2018, in Phoenix, Ariz., after which they were ranked based on their submissions and presentations.
The problem was designed by a group of industry professionals who each year select a current real-world problem in the rail industry. The teams were asked to design a train prediction system to give passengers real-time updates. They were given two months of data – about 10 million data points – showing actual train performance over that time. They also had access to the planned timetable for that same period, historical time performance, the staffing schedule, the maximum allowable speeds for the trains, the number of tracks and weather conditions, including rain, temperature and wind speed.
“The provided dataset was an example of Big Data,” Meidani said. “The problem needed cutting edge methodology – machine learning, data mining, and predictive analytics. These problem solving competitions are great opportunities for industry to bring to the table their needs and challenges and see how the state-of-the-art methodology can solve their problems.”
For students, it’s a learning opportunity and a confidence booster, he said.
“The competition gave me the chance to apply some of my machine learning knowledge to solve a practical challenge,” Alemazkoor said. “I hope railroad companies, including Netherlands Railways that designed the competition, will be able to use our approach to provide passengers more accurate estimations of trains’ arrival times.”
Nabian agreed, saying the chance to work on real-world data made him “feel valuable to a real-world industry” and as such was more satisfying than working on a virtual problem.
The team won a plaque and $1,000, and they have been approached by industry to further develop the prediction system.
A CEE team won first place in the same competition in 2015, which was focused on accurate prediction of track defects. That team include Alemazkoor, Meidani and Senior Research Engineer Conrad Ruppert.
For more information about the award and INFORMS, please visit https://connect.informs.org/railway-applications/awards/problem-solving-competition/result.