Clustering in point processes on linear networks using nearest neighbour volumes

Area of Science:

• Spatial statistics
• Network analysis
• Geographic information systems (GIS)

Background:

• Traditional spatial clustering methods are designed for planar spaces and struggle with the unique geometry of linear networks.
• Point process analysis on networks presents challenges in data visualization and property interpretation.
• Existing methods do not adequately address the detection of localized high-density point clusters within linear network structures.

Purpose of the Study:

• To introduce a novel method for detecting clusters of points specifically within linear networks.
• To adapt existing point process classification approaches for network-based spatial data.
• To identify and analyze regions of increased point density on linear networks, particularly for applications in road safety.

Main Methods:

• The study extends a classification approach for point processes in spatial contexts to linear networks.
• It utilizes the distribution of Kth nearest neighbor volumes to identify regions of increased point density.
• The method is designed to distinguish overlapping point processes within the same linear network.

Main Results:

• The novel method successfully detected distinct clusters of high-density points in road segments prone to severe traffic accidents in Bogota and Medellin.
• Identified accident clusters were predominantly located on major arterial roads with high traffic volumes.
• Low-density point areas corresponded to locations with fewer accidents, likely due to lower traffic flow or other safety factors.

Conclusions:

• The developed method provides an effective tool for identifying high-risk accident locations on road networks.
• Findings offer valuable insights for urban planning and targeted road safety management strategies.
• The approach demonstrates the utility of network-based spatial analysis for understanding accident patterns and improving public safety.