ScaleFC: A scale-aware geographical flow clustering algorithm for heterogeneous origin-destination data

Abstract

Exploring the cluster pattern of geographical flow facilitates the understanding of the underlying process of geographical phenomena among spatial locations. Despite recent advancements in identifying flow clusters, challenges remain when handling flow data with uneven length, heterogeneous density and weak connectivity. To solve the issues, this study proposes a Scale-aware Flow Clustering algorithm (ScaleFC). It identifies flow clusters of arbitrary lengths by employing an analytical scale to generate an adjustable neighborhood range of each flow. Meanwhile, inspired by the idea of boundary-seeking clustering, ScaleFC introduces partitioning flows to identify flow clusters with different densities, and separate the weakly-connected clusters. To validate the effectiveness, we compared ScaleFC with three mainstream baselines, i.e., AFC, FlowLF and FlowDBSCAN, on six synthetic datasets. The results presented that ScaleFC can accurately identify the clusters with complex structures, achieving an average accuracy improvement of 27 %, 17 %, and 15 % over the three competitors, respectively. The application on bike-sharing data with 16,140 flow pairs from Shanghai City demonstrated that ScaleFC is capable to capture both long-distance and short-distance movements, thereby providing a more comprehensive understanding to multi-scale human mobility patterns in geographical space.