ORCID Identifier(s)

0000-0001-6350-0342

Graduation Semester and Year

Spring 2025

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Civil Engineering

Department

Civil Engineering

First Advisor

Kate Hyun

Second Advisor

Stephen Mattingly

Third Advisor

Nelson R. Gomez-Torres

Fourth Advisor

Shouyi Wang

Abstract

Paratransit services play a vital role in supporting the mobility of older adults and individuals with disabilities, yet they experienced major disruptions during the COVID-19 pandemic and continue to face ongoing operational issues such as trip cancellations. This dissertation explores these challenges through three distinct analyses using data from Arlington, TX. The first chapter investigates how COVID-19-related factors (policies, cases, and vaccination) and socio-demographics influenced paratransit usage from 2019 to 2021. Applying SARIMAX and Poisson regression models, the study finds a substantial drop in ridership—41% in 2020—and a shift toward essential medical travel. It also reveals that restrictive policies, such as workplace closures, had a stronger effect on demand across all trip types than case counts or vaccination rates. Demographic characteristics consistently shaped usage patterns. The second chapter analyzes usage trends between 2019 and 2022 using time-series clustering. Four distinct user groups were identified, each with different trip frequencies and recovery patterns following the pandemic. While high-frequency riders largely resumed their travel, moderate users only partially returned, and over half (55%) discontinued service altogether—underscoring the need for user-specific service strategies. The third chapter focuses on improving operational efficiency by predicting trip cancellations and no-shows through machine learning. A hierarchical classification approach, combined with strategies to address class imbalance (such as undersampling, oversampling, and class weighting), significantly outperformed baseline models. Random Forest with class weighting yielded the best overall results, while SVM was especially effective for identifying no-shows. These predictive tools can help agencies allocate resources more efficiently, reduce unused capacity, and enhance reliability. Together, these studies offer critical insights into paratransit system resilience during crises and provide data-driven methods to improve service planning, offering actionable guidance for future disruptions and better service delivery to vulnerable populations.

Keywords

Paratransit, COVID-19, Travel Behavior, SARIMAX, Time-Series Clustering, Machine Learning, Prediction

Disciplines

Civil Engineering | Transportation Engineering

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Available for download on Tuesday, May 05, 2026

Share

COinS