Bike Network Flow Prediction

Recent network analysis research has made remarkable progress on studying complex networks in a social context with machine learning techniques. One example of such a complex network is the system data of the New York Citibike bike sharing program, which consists of the data from 49 million bike trips in New York City since July 2013. By inspecting the dataset, we find out that significantly more demand could be satisfied if we were able to predict the bike flow between stations and rebalance the bike supplies accordingly. In this paper, we formalize this task as predicting the change in number of bikes at each station, and develop algorithms to predict this change based on snapshots of the network from the past. We find that clustering the network into smaller parts results in more accurate predictions, and propose a novel algorithm for rebalancing the network based on the prediction results from our model.