Identifying mode of transport for partial trips - when analyzing movement using mobile network data

Important results from the project

The project did not achieve the goal of creating a self-learning AI model for mode identification. The model hypothesis was found to identify statistical artifacts rather than real speed profiles. An alternative method was developed with journey planners as support for annotation, which is used commercially and is being evaluated by the City of Stockholm. A follow-up project is initiated by Swedish Transport Administration and we identified valuable insights into journey chain-based analysis.

Expected long term effects

Insights and new collaborations created in the project are expected to contribute to improved traffic planning through more cost-effective and reliable methods for mode identification. The developed method with support from travel planners can replace or complement traditional travel habit surveys that have declining response rates. The insights into travel chain-based analysis enable a better understanding of travel patterns, which is crucial to achieving the goals of sustainable mobility.

Approach and implementation

The project was largely on schedule, but unexpected challenges arose during validation when the model was found to produce statistical artifacts rather than identify true speed profiles. This led to a three-month extension to explore alternative methods using trip planners as training data for a KNN model. Despite the challenges, the collaboration between the project partners worked well, providing valuable lessons learned and paving the way for continued development in the new project.

External links

Drive Swedens hemsida