Leveraging Mobility Data to Simulate Trip Chains in Traffic Systems Using a k-anonymised Bayesian Network

The transportation sector plays a crucial role in reducing global greenhouse gas emissions. In this context, simulation models enable virtual testing of intelligent mobility solutions and prediction of mobility behaviour under various conditions. Improving the accuracy of these models requires insights from real-world data. This need often conflicts with anonymisation and data protection requirements, particularly for small data sets. Privacy-preserving methods are therefore essential to safeguard data donors’ privacy while leveraging real mobility data to enhance simulation models.This work introduces a method for extracting information about individual mobility behaviour that ensures data privacy and integrates the data into a detailed traffic simulation model. A Bayesian network is constructed to capture dependencies between trip variables such as trip purpose and start time. The network graph is learned using a genetic algorithm applied to k-anonymised data. Conditional distributions derived from these dependencies are then used to generate synthetic, individual trip chains that serve as input for a traffic simulation in the open-source software SUMO. As a result, synthetic trajectory data are produced, enabling the study of different mobility behaviours. The two main contributions of this work are the method to build an anonymised detailed trip chain model and the usage of this model as an input for a traffic simulation model.The proposed method has been validated and tested with real mobility data and can be extended in the future by incorporating additional data sources and assumptions to further improve simulation accuracy.