Incorporating parametric action decision trees in computational process models of activity-travel behavior: theory and illustration

As an alternative to utility-maximizing nested-logit models, Albatross uses decision trees to predict the activity-scheduling decisions of individuals and households. The decision trees are derived from activity-diary data and are able to account for discontinuous and non-linear effects of independent variables on choice variables. A potential disadvantage of rule-based models is that the sensitivity of predictions of travel demand may be reduced. To overcome this problem and combine the specific strengths of the rule-based and parametric modeling approaches, the authors have developed a hybrid approach referred to as parametric decision trees. The paper describes the approach and results of incorporating the extended decision trees in Albatross to improve the sensitivity of the model for travel-time and travel-costs scenarios.