Activity-Based Travel Demand Modeling Framework FEATHERS: Sensitivity Analysis with Decision Trees

Abstract

The decision tree technique is increasingly applied in activity-based travel demand modeling. It has the strength of representing the full complexity of interactions between different variables. However, this complexity often hinders an interpretation of the relative effects of these variables on the activity travel choice. In this study, a sensitivity analysis is performed on decision trees in FEATHERS, an activity-based microsimulation modeling framework, with the purpose of quantitatively measuring the relative effect of input variables involved in the given decision trees on the choice variable. The local and global sensitivity analysis approaches are investigated: ( a) a one-at-a-time approach that predicts the choice frequency distribution by varying selected input variables one after another and keeping all other variables as observed and ( b) the improved Sobol method, which evaluates the effect of an input variable while all other variables are varied as well. With the application of these two approaches to two representative decision trees concerning work-related activity (i.e., commute trip) choice and transport mode choice for work-related activities in the FEATHERS framework, consistent results concerning the key input variables for these two decision trees are derived, and some extra insights are gained from each of the approaches.