Machine Learning and Big Data-Based Approaches for Quality Freeway Volumes

Adrian Cottam, University of Arizona


Traffic volumes are essential for the operation of Intelligent Transportation Systems (ITS) and for use in performance metrics for Transportation Systems Management and Operations (TSMO). Maintaining high quality freeway volumes is paramount to ensure that freeway ITS solutions, such as ramp meters, operate smoothly. Furthermore, freeway performance metrics need quality volume data to provide an accurate portrayal of real-world conditions. 
This dissertation focuses on improving the quality of freeway volumes, as well as expanding their spatial availability. The dissertation is comprised of three components. In the first component, loop detector data and crowdsourced data are used to develop a machine learning agent network to estimate volume for faulty loop detectors. In the second component, crowdsourced data and recurrent neural networks are used to estimate freeway volumes at locations where no loop detectors are available. In the third component, crowdsourced data is used to estimate freeway volumes at locations where no loop detectors are available for several different types of models using spatial embedding. 
These methods can provide benefit for transportation professionals, as machine-learning volume data estimation can allow them more time to repair faulty detectors, as well as reducing the need for freeway sensors in some areas.

Project Details

Project Type:
Project Status:
End Date:
January 31,2024
UTC Grant Cycle:
NITC 16 Dissertation Fellowships
UTC Funding:

Other Products

  • Volume Estimation Machine Learning Agent Network: Imputing Missing Data for Failed Freeway Traffic Sensors (PRESENTATION)