Authors: Yongsung Lee*, Georgia Institute of Technology - Atlanta, GA, Bumsoo Lee, University of Illinois, Urbana-Champaign
Topics: Transportation Geography, Quantitative Methods, Urban and Regional Planning
Keywords: public transit ridership decline, ridehailing, machine learning,
Session Type: Paper
Presentation File: No File Uploaded
Despite substantial investment in public transit systems in the United States in the 2000s, most American cities (except a few exceptions such as Seattle and Houston) have witnessed flat or declining transit ridership in the 2010s. While economic recovery since the last recession (i.e., increasing incomes and cheap gas prices) certainly led to rebounding vehicle miles traveled and (associated) reduction in transit use in the U.S., the overall declining customer base of the US transit systems is still puzzling given the continuing political and financial support for them. In response, researchers and analysts started to examine the effects of various factors behind this trend. Those factors include teleworking, increasing vehicle ownership among low-income and immigrant households, growing adoption of alternatives (e.g., ridehailing services and bike sharing systems), gentrification of transit-rich neighborhoods, and of course, service reliability issues.
In this study, we estimate synergistic effects of economic factors and land use attributes on transit ridership of large urbanized areas (UZA) in the U.S. from 2002 to 2017. We employ Gradient Boosting Decision Trees (GBDT) on a continuous dependent variable (i.e., log-transformed monthly unlinked trips). GBDT is an ensemble machine learning model, which develops a stronger predictive model (i.e., a model producing more accurate prediction) from a weaker model at each iteration. In sum, this study contributes to the literature on the synergistic effects of price-oriented policies and land-use planning for the promotion of sustainable transportation.
To access contact information login