ORCID Identifier(s)

ORCID 0000-0002-5037-564X

Graduation Semester and Year

Spring 2024

Document Type


Degree Name

Doctor of Philosophy in Civil Engineering


Civil Engineering

First Advisor

Dr. Melanie L Sattler

Second Advisor

Dr. Kate Hyun

Third Advisor

Dr. Victoria Chen

Fourth Advisor

Dr. Arpita Bhatt


Telecommuting, increasingly prevalent in today's workforce, has gained significant momentum amid the pandemic-induced shift towards remote work. It altered the travel behavior of workers and their travel patterns. This evolution prompts a reevaluation of worker travel behavior, necessitating adjustments in travel demand modeling to optimize transportation planning and service provision. Using the commuter’s data from a typical pre-pandemic workday to model the travel demand of weeks or years in the future is outdated. In designing transportation systems for the future, Metropolitan Planning Organizations (MPOs) will thus need to account for change in work trips due to telecommuting, and how this impacts mobility, air quality, and public health. A significant need exists for a post-pandemic model for MPOs to forecast telecommuting so that it can be incorporated into Travel Demand Models (TDM) to forecast future congestion and need for additional transportation facilities.

The overall goal of the project was to develop post-pandemic Telecommuting Expectation Models (TeEMs): Choice Model and the Frequency Model, for MPOs nationwide. This study aimed to use supervised machine learning algorithms to predict the choice and frequency of an individual adopting telecommuting using post pandemic data. MPOs can use the TeEMs models to incorporate telecommuting into their travel demand modeling (TDM) and emissions modeling to understand community-level mobility impacts of telecommuting. The models will offer valuable insights for traffic management agencies, enabling them to make informed decisions regarding traffic flow and congestion management. City planners and developers can use them to adapt urban planning strategies to cater to the changing needs of telecommuters. The data utilized to develop the models was derived from responses to the national survey "COVID-19 and the Future Survey.” Supervised machine learning algorithms were used to develop the Choice and Frequency Model. Choice Model is a binary classification model which classifies a worker into commuter or telecommuter, and Frequency Model is a multiclass classification model, which classifies the telecommuters into 5 categories (1 day, 2 days, 3 days, 4 days and 5 days) based on frequency of telecommute adoption in a week. Both models used socio-demographic attributes that are nationally available like age, gender, education, household income, household size, No. of vehicles, children and employed adults in household, employment category, commute time and commute distance to make predictions. Decision trees, Random Forests, Support Vector Machines (SVM), Logistic Regression, Gradient Boosting, AdaBoost and XGBoost algorithms were trained to find the best model. The the main steps followed in model building process included data preprocessing, exploratory data analysis, feature selection, data splitting, model selection, model training, model evaluation, hyperparameter tuning, prediction and documentation.

The AdaBoost model showed the best performance among the algorithms used to predict an individual’s choice of telecommute adoption. The confusion matrix revealed that out of 270 instances, the model correctly classified 77 instances of Telecommuters and 88 instances of Commuters but misclassified 20 instances of telecommuters as commuters and 32 instances of commuters as telecommuters. Additionally, the model achieved an accuracy of 76%, precision of 73.3%, recall of 81.5%, and F1-score of 77.2%. These metrics indicate that while the model performs well overall, there is room for improvement in correctly identifying instances of Class B. Age of the employee emerged as the predominant factor influencing their decision to adopt telecommuting. Commute distance, followed by commute time, number of employed adults in the household, and number of vehicles in the household, also influenced choice to adopt telecommuting.

The Gradient Boost model showed better performance for accuracy and recall, while AdaBoost exhibited better prediction and F1score among the algorithms explored to predict the frequency of telecommute adoption. While the frequency model correctly predicted most of the majority class (i.e. telecommuting 5 days) instances correctly, it had difficulty predicting the minority class, especially telecommute frequency 1 day and 4 days. After applying Synthetic Minority Oversampling Technique, the AdaBoost model could only achieve an accuracy of 46.3%, precision of 43.5%, recall of 46.3%, and F1-score of 44%, indicating that the model still struggles to correctly classify instances from the minority class. This could suggest that the underlying patterns in the minority class are complex or difficult to capture, or that the chosen features or attributes may not be suitable for effectively discriminating between classes. The Frequency Model with additional attributes (attribute set 5) performed better with an accuracy of 55%, precision of 52%, recall of 55%, and F1-score of 53%, however its cannot be used by MPO's now, as these added variables are not available in ACS or NHTS dataset. The model, however, proves our hypothesis of need of more attributes (other than socio-demographic factors) for the Frequency Model.

The TeEMs Choice Model was demonstrated by a case study in the Dallas Fort Worth Metroplex using the 2017 National Household Travel Survey (NHTS) data. The demonstration of Choice model estimated that around 34% of the total sample population in DFW are telecommuters. The findings indicate a notable rise in telecommute adoption within the Metroplex, increasing from 13.5% during pre-pandemic time to 34% in the post pandemic. The scenario analysis predicted 20.5% change in the home-based work trip due to telecommuting in the Metroplex The large increase in telecommuters suggests that it is very essential to update the existing travel demand models to make informed transportation planning and policy-making.

Subsequently, a Transportation Impact Scenario Analysis was conducted to examine telecommuting's effects on transportation, particularly changes in vehicle miles traveled (VMT) and Vehicle Hours Travelled (VHT), accounting for rebound effects pre- and post-pandemic. The Transportation Impact Scenario Analysis evaluated six distinct post pandemic scenarios against a prepandemic baseline to assess the impact of telecommuting adoption on regional transportation dynamics. Scenario 1 evaluated a 15% reduction in HBWT and 9.3% rebound trips, resulting in reduction of 6 million VMT/day (2.65%). Scenario 2, 3, 4 and 5 evaluated 20% trip reduction due to telecommuting and various rebounds ranging from 10.9 to 13.4%. On average, a reduction of 8.3 million VMT per day (3.6%) was observed in these scenarios. The final scenario evaluated 35% telecommute adoption, which was very close to 34% which was the telecommute rate in DFW according to the TeEMs demo. Even with 20.5% rebound trips, VMT decreased by 15.9 million per day (6.9%).

The Vehicle Hour Travelled (VHT) also showed a trend similar to VMT. Scenario 1 showed a reduction of 0.38 million VHT per day (5.65%) from pre-covid s. Scenarios 2, 3, 4, and 5 exhibited an average reduction of 7.5%, equivalent to approximately 0.5 million VHT per day. Meanwhile, Scenario 6 showed a total reduction of 0.9 million VHT per day (13.4%).

The findings suggest that telecommuting has the potential to reduce VMT and VHT. The demonstration of TeEMs Choice Model for DFW suggests that telecommuters have more than doubled post-pandemic. With the growing population influx to the Metroplex, promoting telecommuting could serve as a valuable strategy to alleviate congestion.


Telecommuting, Post-COVID telecommute, Machine learning models, Telecommute choice and freequency


Civil and Environmental Engineering | Civil Engineering | Environmental Engineering | Transportation Engineering


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

Available for download on Thursday, May 28, 2026