CF-Cluster: Clustering Bike Station Based on Common Flows

摘要

Along with the rapid development of green travel of city bike sharing, how to mine moving patterns from dataset of sharing bike have gradually become hot point of bike sharing research (e.g., bike scheduling, city computing, and so on). Stations clustering is the base of these research directions. Existing literature was clustered the stations by their scalar data, such as, the location, the number of bike lent, the number of bike returned, and so on. Obviously, these clusters didn't own similar features of bicycle flows because of no taking the relations between stations into account. In this paper, we propose an algorithm of clustering analyses, called Cluster analysis based on Common Flow (CF-Cluster for short), based on similar relations between stations. In CF-Cluster, the clusters are defined as the station subset, in which the ratio of common relations (Common Flow Ration) is exceeds the threshold. According to the feature of common flow in subset, CF-Cluster divides into two phases. The first is to discovering candidate station subsets through the idea of Apriori, which is classic algorithm of association rules. The second phase is to eliminates overlapped clusters in candidate subsets to obtain station clusters. Finally, Empirical evaluation proves that our algorithm owns availability and effectiveness. Moreover, scale-up experiments show the affects in the number and size of clusters.