The application of machine learning and AI-based methods in traffic and transportation data analysis has revolutionized the way we understand and manage urban mobility. These advanced technologies enable the extraction of meaningful insights from massive datasets, offering a more nuanced understanding of traffic patterns, predicting congestion, and optimizing transportation systems. Machine learning models, such as neural networks and decision trees, can predict travel times, identify traffic anomalies, and recommend efficient routes. AI algorithms excel in processing real-time data from diverse sources, including sensors, GPS, and social media, to enhance traffic management and provide timely information to commuters. Additionally, these methods play a crucial role in developing smart transportation systems, such as autonomous vehicles and intelligent traffic signal systems, contributing to the creation of more efficient, sustainable, and resilient urban transportation networks.
In coordination with cutting-edge science and data analysis methods, I utilized machine learning and AI-based methods in my research. The showcases are supervised and unsupervised methods for outlier detection and classification of trajectory sample points in ANT-2020, Trans. Res. Part C-2021, and Trans. Res. Part C-2022.
Recently, while mentoring a master's thesis, which produced a TRB-2024 paper, we applied Reinforcement Learning (RL) to reconstruct trajectories at the lane level. The feasibility conditions of each trajectory are used to define reward functions in two formulations: one based on single-agent RL (SARL) and the other based on multi-agent RL (MARL).
Recently, I was involved in collaboration with experts from Utah State University regarding the severity of pedestrian crashes. The research is highlighted not only by its contribution to safety analysis but also in its methodology, which consists of a comparison of Machine Learning Methods (Stacking Ensemble Model and TabNet) versus discrete outcome models (Ordered Probit).
(Paper: TRB-2024)