Travel demand generation using Bayesian Networks: an application to Switzerland

Abstract

Thanks to their ability to simulate the travel behavior at the individual scale, agent-based models have gained popularity over the last years. These models are data-intensive, with regards to transport supply and demand. In particular, a detailed description of the population and its travel behavior is required. Bayesian Networks (BNs) are directed acyclic graphs representing joint probability distributions. They have recently been employed for population synthesis and daily activity patterns generation in studies showing that BNs effectively capture the causality links existing between variables and are easily interpretable. Moreover, given their flexible structure, BNs can be adapted for situations in which data from various sources is combined. In this paper, our goal is to estimate a BN for both population and activity pattern synthesis in Switzerland. We evaluate the performance of this approach compared to the statistical matching algorithm using aggregated and disaggregated metrics. In particular, we show that understanding the dependency structure linking the population characteristics and its mobility behavior is key to generate representative synthetic agents and daily activity patterns. This study is a contribution towards the development of interpretable, flexible and behaviorally rich travel demand generation models.