Development and Application of TeEMs (Telecommuting Expectation Models): Predicting Post-Pandemic Telecommuting Choice and Frequency Using Machine Learning Models

Abstract Description: Telecommuting, increasingly prevalent in today's workforce, has altered the travel patterns of workers, necessitating adjustments in travel demand modeling to optimize transportation planning and services. This study leverages post-pandemic data from the “COVID-19 and the Future Survey,” a collaboration between Arizona State University and the University of Illinois Chicago, supported by the National Science Foundation. Using supervised machine learning algorithms, two models were developed: a binary Choice Model to classify workers as commuters or telecommuters and a multiclass Frequency Model to categorize telecommuters by weekly frequency (1–5 days). Both models use socio-demographic data such as age, gender, education, income, household details, commute time, and distance. Machine learning algorithms, including Decision Trees, Random Forests, SVM, Logistic Regression, Gradient Boosting, AdaBoost, and XGBoost, were tested. AdaBoost outperformed others in predicting telecommuting adoption, offering valuable tools for managing traffic flow and congestion.

A case study using the 2017 National Household Travel Survey data in the Dallas-Fort Worth Metroplex demonstrated the TeEMs Choice Model, showing telecommuting adoption increased from 13.5% pre-pandemic to 34% post-pandemic. Scenario analysis predicted a 20.5% reduction in home-based work trips due to telecommuting, impacting vehicle miles traveled (VMT) and vehicle hours traveled (VHT). In Scenario 1, a 15% HBW( home based work) trip reduction and 9.3% rebound trips led to a 2.65% daily VMT reduction (6 million miles). Scenarios 2–5 modeled 20% trip reduction with 10.9–13.4% rebounds, averaging 8.3 million fewer VMT per day (3.6%). Scenario 6, reflecting 35% telecommuting(close to 34% which was the telecommute rate in DFW according to the TeEMs demo), showed a 6.9% daily VMT reduction (15.9 million miles), despite 20.5% rebound trips. VHT trends mirrored VMT reductions. Scenario 1 saw a 5.65% drop (0.38 million hours/day). Scenarios 2–5 averaged a 7.5% reduction (0.5 million VHT/day), while Scenario 6 achieved a 13.4% decrease (0.9 million VHT/day).These results suggest telecommuting can significantly reduce VMT and VHT. With rising Metroplex population, promoting telecommuting could help mitigate congestion effectively.