Bike-Sharing Prediction System

Bike Sharing System is a dynamic network. This paper proposes a method to balance the network and allocate the bikes in each station to avoid the imbalance happening. Real-time monitoring takes too much time to reallocate the bikes if an imbalance has occurred so it cannot tackle this problem well. And most of sharing systems cannot focus on the influenced factors and overlook assuring the creditability of the prediction. In this paper, we propose a hierarchical prediction model to predict the number of bikes. It mainly contains the following parts. First, we propose a clustering algorithm to cluster bike stations into groups using Gaussian Mixture Model (GMM). Second, gradient Boosting Regression Tree(GBRT) is adapted to predict the entire triffic. Third, we predict the proportion across clusters and the inter-cluster transition using a multi-factor-based inference model. Finally, we adapte Geo-Space Contrary Prediction Model to compare with the same period prediction datasets to improve the results. Based on Citi Bike system data in NYC, from Apr. 1^st, 2014 to Sept. 30th, 2014 and the influenced factors, our model outperforms baseline approaches and can be applied to various geograph scene.