Data Driven Methods to Improve Traffic Flow and Safety Using Dimensionality Reduction, Reinforcement Learning, and Discrete Outcome Models

Author

Kazi Redwan Shabab, University of Central FloridaFollow

Keywords

Adaptive Traffic Intersection, Reinforcement Learning, Traffic Safety, Dynamic mode decomposition, Discrete Outcome Models

Abstract

Data-driven intelligent transportation systems (ITS) are increasingly playing a critical role in improving the efficiency of the existing transportation network and addressing traffic challenges in large cities, such as safety and road congestion. This dissertation employs data dimensionality reduction, reinforcement learning, and discrete outcome models to improve traffic flow and transportation safety. First, we propose a novel data-driven technique based on Koopman operator theory and dynamic mode decomposition (DMD) to address the complex nonlinear dynamics of signalized intersections. This approach not only provides a better understanding of intersection behavior but also offers faster computation times, making it a valuable tool for system identification and controller design. It represents a significant step towards more efficient and effective traffic management solutions. Second, we propose an innovative phase-switching approach for traffic light control using deep reinforcement learning, enhancing the efficiency of signalized intersections. The novel reward function, based on speed, waiting time, deceleration, and time to collision (TTC) for each vehicle, maximizes traffic flow and safety through real-time optimization. Finally, we introduce a mixed spline indicator pooled model, an approach for multivariate crash severity prediction, addressing the limitations of previous models by capturing temporal instability. It carefully incorporates additional independent variables to measure parameter slope changes over time, enhancing data fit and predictive accuracy. The developed models are estimated and validated using data from the Central Florida region.

Completion Date

2023

Semester

Fall

Committee Chair

Eluru, Naveen

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Civil, Environmental, and Construction Engineering

Degree Program

Civil Engineering

Format

application/pdf

Identifier

DP0028090

URL

https://purls.library.ucf.edu/go/DP0028090

Language

English

Release Date

December 2024

Length of Campus-only Access

1 year

Access Status

Doctoral Dissertation (Campus-only Access)

Campus Location

Orlando (Main) Campus

STARS Citation

Shabab, Kazi Redwan, "Data Driven Methods to Improve Traffic Flow and Safety Using Dimensionality Reduction, Reinforcement Learning, and Discrete Outcome Models" (2023). Graduate Thesis and Dissertation 2023-2024. 74.
https://stars.library.ucf.edu/etd2023/74