Exploiting Interpretable Patterns for Flow Prediction in Dockless Bike Sharing Systems

Unlike the traditional dock-based systems, dockless bike-sharing systems are more convenient for users in terms of flexibility. However, flexibility these comes at cost management and operation complexity. Indeed, imbalanced dynamic use bikes leads to mandatory rebalancing operations, which impose a critical need effective bike traffic flow prediction. While efforts have been made developing prediction models, existing approaches lack interpretability, thus limited value practical deployment. To this end, we propose an Interpretable Bike Flow Prediction (IBFP) framework, can provide with interpretable patterns. Specifically, by dividing urban area into regions according density, first model spatio-temporal flows between graph regularized sparse representation, where Laplacian is used as smooth operator preserve commonalities periodic data structure. Then, extract patterns from using subspace clustering representation construct base matrices. Moreover, be predicted matrices learned parameters. Finally, experimental results on real-world show advantages IBFP method sharing systems. In addition, interpretability our pattern exploitation further illustrated through case study provides valuable insights analysis.