Towards activity-based demand modelling for the Greater Oslo Area - Using machine learning to predict travel mode choice and activity plans

This report documents the R&D efforts conducted as part of the PRELONG project between 2022 and 2024, focusing on generating activity plans for a synthetic population in the Greater Oslo Area. At the core of our approach are two neural network models that predict the characteristics of all trips made on a typical weekday. These machine learning models are trained on travel survey data from RUTER-MIS (2017–2024) and are applied to population and commuting data. Among the many potential applications of these datasets are agent-based simulation models that use this data as input. Furthermore, additional analyses explore the influence of level-of-service variables on the prediction of travel mode choices, highlighting certain challenges associated with a purely data-driven approach.