Predicting Mode Choice on Urban Work Trips by Non-Private Vehicles

In the great metropolitan area, under the low carbon transportation framework and imbalanced job-housing situation, some people choose to travel through non-private vehicles. Some governments would endorse such behavior by all means as congestion gets worse. For this study, the 2017 National Household Travel Survey is selected, targeting the individuals in urban areas who travel by walking, bicycling, public transit, and taxis for their work trips. A discrete choice (multinomial logit), unsupervised (k-means clustering and principal component analysis), and supervised (naïve Bayes classifier and random forest) models are fitted to test the predictive performance, with the vision to deeply master the system partial demand and design a better traveling system for citizens indirectly. The results show that unsupervised methods for predicting work trip mode choices needs to be further discussed while some supervised learning methods could be considered as a promising reference for predicting travel modes.