Data Mining Approach to the Work Trip Mode Choice Analysis in the Chicago Area

Discrete Choice Methods have brought attention to the study of the travel mode choice behavior in both theoretical and practical areas of transportation planning and modeling. Unlike discrete choice models that impose pre-defined probability distributions on choice probabilities, data mining approaches view the travel mode choice as a pattern recognition problem whereby the travel choices can be identified by a combination of explanatory variables. As such, data mining techniques have enjoyed increasing applications in agent-based modeling. This paper examines the capability of a promising machine learning algorithm, Class Association Rules (CAR), for the estimation of the work trip modal choice using Chicago Area Transportation Studies (CATS) 1990 Household Travel Survey (HTTS) as the case. The purpose of the paper is to investigate the advantages, disadvantages and applicability of CARs for the study of mode choice behavior. It reveals that the CAR is a useful tool for building a powerful mode choice model, with the overall accuracy reaching up to 93% for the data set used in this study. Unlike some of other mining methods, the rules extracted by CAR are easy to interpret and provide insights into the travel behavior from a perspective that is different from statistical models. The example presented in this paper also illustrates one of the key advantages of CAR over discrete choice models, which is the flexibility of the model specification.