Latent class choice model with a flexible class membership component: A mixture model approach

This study presents a semi-nonparametric Latent Class Choice Model (LCCM) with flexible class membership component. The proposed model formulates the latent classes using mixture models as an alternative approach to traditional random utility specification aim of comparing two approaches on various measures including prediction accuracy and representation heterogeneity in choice process. Mixture are parametric model-based clustering techniques that have been widely used areas such machine learning, data mining patter recognition for classification problems. An Expectation-Maximization (EM) algorithm is derived estimation model. Using different case studies travel mode behavior, compared discrete basis parameter estimates' signs, value time, statistical goodness-of-fit measures, cross-validation tests. Results show improve overall performance by providing better out-of-sample addition representations without weakening behavioral economic interpretability models.