Parametric Action Decision Trees: Incorporating Continuous Attribute Variables into Rule-Based Models of Activity-Travel Behavior

Rule-based models, such as decision trees, are ideally suited to represent discontinuous and non-linear effects of independent variables on activity-travel choice behavior. At the same time, however, the models require that continuous attributes, such as for example travel time and travel costs, are discretisized which may decrease the sensitivity of predictions for travel demand measures that involve these attributes. To overcome this problem and combine the specific strengths of the rule-based and parametric modeling approaches, this paper introduces a hybrid approach. The so-called parametric action decision tree (PADT) replaces the conventional action assignment rule of the decision tree by a logit model or any other parametric discrete choice model. The model includes a set of alternative specific constants to take the impact of leaf node membership of a case into account in addition to terms for the continuous attributes. As an illustration, it is shown how the approach can be used to incorporate travel costs sensitivity in Albatross - a computational process model of activity scheduling. Using the work activity as a test case, the results suggest that the enhanced, hybrid model can reproduce realistic price elasticities of travel demand.