Spatial Contiguity-Constrained Hierarchical Clustering for Traffic Prediction in Bike Sharing Systems

The critical problem in managing a bike sharing system (BSS) is to solve the imbalance in the number of available bikes by stations and times, which negatively affects the users' riding experience. To address this issue, many BSS operators rearrange bikes using a fleet of trucks. Moreover, the effectiveness and efficiency of rebalancing operations heavily rely on accurate traffic prediction in BSS. Hence, many researchers have developed methods to enhance the prediction performance. One approach to increase the accuracy is to hierarchically predict demand from higher level (cluster or system) to station level. In the hierarchical prediction frameworks, the cluster-level to predict the demand of the total usage of groups of stations is introduced and it is crucial to partition stations into clusters that can increase the stability of actual usage and prediction. Therefore, the present work proposes a new spatial contiguity-constrained hierarchical clustering algorithm for BSS. The proposed algorithm is based on hierarchical clustering, which is deterministic, and uses the hourly proportions of check-outs and check-ins to define the temporal usage pattern of each station that does not require cluster assignment. Thus, different from existing methods, our proposed algorithm is deterministic and fast. In addition, the proposed algorithm is proven to be superior to other clustering methods in terms of traffic prediction at both the cluster and station levels.