Project information
- Category: Machine Learning
- Data: NDW loop detector data & NWB network data
- Project: EU Project SETA
- Methods: Feature Engineering, Image Processing, Clustering, Active Shape Models, 3D Maps
- Technologies: MATLAB, Python, pandas, geopandas, scikit-learn, opencv
- Journal: Transportation Research Part C
Project details
This research developed an innovative method to simplify and accurately capture network traffic patterns in large-scale networks. Using 'congestion pockets' as 3D shapes, we turn them into interpretable feature vectors via 2D projections. These vectors can predict travel times and reveal congestion patterns. In testing, we achieved a 44% accuracy improvement over the consensus method for Amsterdam's urban network. When applied to the Dutch highway network, the method showed 93% prediction accuracy. With these compact vectors, we can integrate more data without increasing complexity, promising further improvements.