Destination Prediction for Sharing-Bikes' Trips

Bike-sharing system has been very popular all over the world as it provides benefits like healthy lifestyle and convenience for users. For better dispatching these sharing bikes to the most needed places at any time, a precise prediction of the destination is needed. Unfortunately, existing approaches for destination prediction are used in systems with fixed stations or taxi scenarios. But in our scenario, people can pick up or drop off bikes at any places, which increases the predicting difficulty. In this paper, a data-driven approach is proposed to predict destinations based on large-scale bike trip data. We first formulate destination prediction as a binary classification problem and introduce two different approaches to construct our dataset. After that, different strategies are presented to generate potential candidates and extract multi-view features from historical data. Finally, we train a classifier and returns potential destinations ranked by their probability decreasingly. Experiments conducted on the real-world bike-sharing system dataset demonstrate the effectiveness of the proposed method.