How can we reduce traffic jams during the morning commute?
Mayank K. Vadaliya, doctoral candidate in information technology at the University of the Cumberlands, explores how prediction models could help.
Faculty Bio:
Mayank Vadaliya is a doctoral candidate in Information Technology at the University of the Cumberlands, where his research focuses on machine learning applied to transportation systems. He holds a Master’s in Computer Science from Cleveland State University and is a Full Member of Sigma Xi, the Scientific Research Honor Society. His work on reinforcement learning for adaptive traffic signal control investigates how learning agents can reduce delay and emissions across urban intersection networks.
Transcript:
Every morning, millions of us sit in traffic that seems to appear out of nowhere. One minute the freeway is moving; the next, it’s a parking lot. What if we could see that jam forming before it actually happens?
That’s the question behind my research. I build systems that predict traffic congestion in real time — several minutes before it builds — so drivers and traffic operators have time to react.
Most prediction tools lean on a single model. Some are good at the patterns that repeat, like the daily rush hour. Others are better at sudden, irregular changes — a crash, or bad weather. But no single method handles both well.
My approach combines three kinds of machine learning. First finds patterns across the road network. Second tracks how traffic changes over time. And last improves the final forecast. Together, they cover each other’s blind spots.
I tested this on METR-LA, a public dataset from hundreds of real sensors on Los Angeles freeways. Against standalone models, the hybrid system cut prediction error by more than thirty percent.
Why does that matter? A forecast that arrives a few minutes earlier gives the system time to act — to reroute drivers, to retime signals, to warn commuters before they hit the bottleneck. Congestion costs American
cities tens of billions of dollars a year and adds needless carbon to the air.
Better prediction won’t make traffic disappear. But learning to see the jam before it forms is the first real step toward clearing it.
