Clustering Method for Characterizing Areas of Spatial Networks Based on Degree Mixing Patterns

Abstract

We address a problem of classifying and characterizing spatial networks in term of local connection patterns of node degrees, by especially focusing on property that the maximum node degree of these networks is restricted to relatively small numbers. We mainly consider spatial networks constructed from urban streets by mapping the intersections of streets into nodes and the streets between nodes into links. To this end, we propose a new clustering method for classifying and characterizing a road network automatically into some areas and characterizing them based on the tri-mixing patterns of node degrees and the K-medoids clustering algorithm. The proposed method first computes a feature vector for each node, consisting of the normalized frequency of the ego-centric tri-mixing patterns, then classifies these vector into some areas by the use of the greedy K-medoids clustering algorithm. In our experiments, using the three road networks of three cities in Shizuoka prefecture collected from OpenStreetMap(OSM), we evaluate the characteristic of our method in comparison to a variant method whose feature vectors are computed by the distance from a set of some facility positions. We show that our method can produce the clustering results comparable to those obtained by the variant method without using some extra information like facility positions.