A latent class model with fuzzy segmentation and weighted variables

Latent class models (LCMs) can yield powerful improvements in understanding the travel behaviour over traditional approaches. All the LCMs studies in transportation used discrete choice models for both the choice model and the identification of segment membership. This paper introduces an innovative segmentation methodology for the segment (class) identification model. The method includes a fuzzy segmentation process, which takes into account the varying levels of influence of each attribute on the degree of association with a segment. Five mode choice models were estimated using a data set from a household survey: a multinomial logit model, a nested logit (NL) model, a traditional LCM, a LCM using new segment identification, and a mixed NL. The estimation results indicate that the new segmentation method used for LCM captures heterogeneity differently than the traditional models, with similar likelihood estimates and good prediction results.