Neural network analysis of travel behavior: Evaluating tools for prediction

This article explores the application of neural networks to a behavioral transportation planning problem. The motivation for adding neural networks as a new modeling methodology stems from its apparent relevance to problems requiring large scale, highly dimensional, data analysis, such as travel related behavior. Neural networks provide a tool to analyze the data in which we can model our intuition, and they provide that capability without the complication of having to formalize all the complex causal variables and relationships which other models require. The transportation issue explored, upon which the neural network methodology is tested, is a comparison of travel demand patterns of men and women in Israel. The information base is the Traveling Habits Survey (Central Bureau of Statistics, Israel, 1984, Statistical Abstract of Israel, No. 35) commissioned by the Israel Ministry of Transport; combined with demographic and socioeconomic data of the 1983 Population and Housing Census. As extensive as such surveys are, the neural networks imply that additional categories of data are necessary to predict how these elements relate to travel behavior. This article concentrates on the extent to which neural networks can combine the relative simplicity of aggregate transportation models, with the theoretical advantages and level of detail of disaggregate transportation models, without the latter's complexity. We describe the various directions we took in analyzing complex travel related data with feed forward, backpropagation trained, neural networks.