Uncovering Calendar-Based Structures Through Clustering of Daily Traffic Volume Profiles

Understanding traffic flow patterns is crucial for effective transportation management, urban planning, and congestion mitigation (Boyce 2012). To extract meaningful insights from dense traffic volume data which is usually continuously collected across multiple locations, effective dimensionality reduction techniques are required. This study aims to uncover the recurring temporal patterns and cross-location similarities in daily traffic volume by identifying clusters of days throughout the year that exhibit comparable traffic flow behavior across multiple locations. The study is conducted on a dataset capturing traffic volume across 26 locations for over a year. We first apply Principal Component Analysis (PCA) separately to each location to summarize dominant features in the location-specific daily traffic curves. We then combine these reduced representations across all locations and apply Uniform Manifold Approximation and Projection (UMAP), a nonlinear dimension reduction technique, on them to visualize and identify potential groupings of days with similar daily traffic volume characteristics. The PCA analysis reveals that location-specific daily traffic volume profiles exhibit some consistent structures across days, and UMAP results show that there are clusters of days corresponding to different school calendars, holiday periods and seasonal routines that show similar daily traffic volume patterns across locations. These findings provides insights to forcast fluctuations in traffic usage, hence aid tailoring transportation planning and infrastructure maintenance interventions.

Click report to view more.