Abstract
The activity duration in activity-based approach typically has been analyzed by various types of regression models, which manifest relationships between socio-economic variables for independent variables and activity duration for dependent variable. Among various approaches, the most frequently adopted model is the hazard-based model, which is a parametric approach because it assumes the probability distribution of the dependent variable prior to model estimation. Since the distribution is not usually aware of the assumption of distribution function prior to estimation is sometimes very strong restriction. Especially, when the true distribution has a unique pattern (for example, bimodality shape), we have difficulty in choosing the relevant probability distribution functions. This study aims to develop the activity duration models using kernel density estimator (hereafter ‘KDE’). KDE is a type of nonparametric estimation methods and can construct the probability distribution including some special features, which parametric methods hardly describe. In addition, relationships between travel time and activity duration are also investigated by the bi-variate KDE using travel diary survey data in Seoul, Korea.
