UCF research team earns second place in national threat detection competition

By Cassidy Mills, Staff Writer

October 15, 2025

 

A research team from the University of Central Florida placed second in the 2025 Algorithms for Threat Detection Challenge, a national competition sponsored by the National Geospatial-Intelligence Agency and hosted by Penn State’s Applied Research Lab.

 

The competition, held from May 14 to Aug. 25, challenged teams across the country to build ship routes using limited, anonymous data that showed only the time, location and speed of ships—without any details any identifying details.

 

The goal was to connect individual observations, called posits, to form complete tracks for unknown ships over a 24-hour period in the Gulf Coast and Florida Straits. The data was intentionally sparse and included small random mistakes in position and time to make the task simulate real-world conditions.

 

The UCF team included students and researchers from UCF’s Institute for Simulation and Training (IST) and was led by Hsin-Hsiung “Bill” Huang, an associate professor in the Department of Statistics and Data Science.

 

Team members included students Joshua White, Yuh-Haur “Jackie” Chen and John “Mahlon” Scott. Bulent Soykan, an associate research scientist at UCF IST, oversaw the project.

 

“This achievement is a great showcase of UCF’s strengths in data science, simulation and interdisciplinary research,” Soykan said. “It reflects the team’s hard work, creative thinking and strong collaboration between departments.”

 

Soykan said the team’s strategy was based on a key realization: a single, one-size-fits-all algorithm couldn’t handle the wide range of behaviors seen in real-world maritime traffic. To address this, they built a “mixture-of-experts” hybrid model — a system that functions like an intelligent dispatcher.

 

The system first analyzes each vessel’s movement in real time, identifying whether a ship is moving quickly in open waters, maneuvering in tight spaces, or drifting slowly. Based on that behavior, Soykan said the system routes the data to a specialized algorithm designed for that specific type of movement.

 

“One of the biggest challenges was the massive scale of data required to train the deep learning model,” Soykan said. “Processing years of historical maritime data would have simultaneously.

 

By dividing the workload across processors, the team reduced a task that would normally take 55 days to less than an hour. The accelerated processing speed allowed them to rapidly test and refine their algorithms.

 

“Competitions like this are invaluable,” Soykan said. “They provide a direct, hands-on experience with the kind of complex, messy, large-scale data that defines modern AI and data science challenges.”

 

He added that the team’s hybrid model has promising real-world applications. It could enhance maritime domain awareness for agencies such as the Coast Guard and Department of Defense by automatically detecting unusual ship behavior—such as veering off course — which could signal mechanical issues or suspicious activity.

 

The model also has potential commercial uses, including optimizing shipping routes and managing port traffic. Soykan said the hybrid model provides a foundation for future research, with opportunities to refine its deep learning and classification components and apply it to other domains such as air or ground traffic tracking.

 

“We plan to publish our findings in academic journals and present them at research symposiums,” Soykan said. “This success has certainly energized our research group to tackle even more ambitious projects.”