Mode Choice Analysis And Prediction Of Trip Chaining Behavior

This paper examines the relationship between mode choice of trip chaining and several variables, such as the complexity of trip chaining and the trip maker's characteristics. A multinomial logit model is developed in this paper. Model parameters are estimated using the 2001 National Household Travel Survey (NHTS) data collected from Dane county, Wisconsin, USA. The choice set of the modes are drive alone (DA), shared ride (SR), bus, bike, and walk. With the understanding of travel behavior of trip chaining, it could be easier for agencies to make decision on establishing transportation policy to avoid travel delay. Planners can also have a better understanding about how to improve transportation policy. For example, public transit is seldom used in trip chaining. The reason is probably that the mobility of public transit is not suitable for the travelers. Travelers are more likely to choose drive alone because of its convenience. However, delays will increase if most of the travelers choose drive alone as their mode. If the planners can increase mobility of public transit, improve the connection between each place for public transit, or encourage more travelers to use public transit, they could solve this problem. With deep understanding of the trip chaining behavior, we could also make better predictions on how different types of modes such as automated vehicles or UBER would have impacts when involved in the transportation system. Based on the findings from the 2001 NHTS data, the difference between trip chaining patterns are further examined to see if there is robust growth in trip chaining over time. The 2009 and 2017 NHTS data collected from Wisconsin, USA are used. Besides, the difference between trip chaining patterns over locations are examined as well. Seven cities chosen as the seven smartest cities from the U.S. Department of Transportation are examined in this study. They are Austin, Columbus, Denver, Kansas City, Pittsburgh, Portland, and San Francisco respectively. After examining which factors have statistically significant impacts on mode choice of trip chaining, 10-fold cross validation method is applied to find how well the multinomial logit models are on predicting the trip chaining travel behavior. Besides, the accuracy rate of the neural network model is also computed for the purpose of comparison with the multinomial logit model. Keywords: Tours, Trip chaining, Travel behavior, Mode choice, Multinomial logit model, Machine learning, Neural network model, Ensemble bagging model, 10-fold cross validation