Update 2017/9/15

Dear participants, due to some (potential) data leak, we decided to delete some data in both public and private data sets during the scoring process. This won't affect submission, so you should submit your results which are identical in the previous format. After 0:00 Sept 16 (UTC time), all new scores will be listed at the temporary leaderboard page (link: https://biendata.com/competition/mobike/leaderboard_temp/).  The link will be valid on Sept 16.  We need some time to modify old scores, and combine the old and temporary leaderboard after that. We appreciate all participants who reminded potential leaks to us. If you have any question, please don't hesitate to send an email to support@biendata.com. In addition, the deadline for emerging team is at 0:00 Sept 18 (UTC time).

Two anonymous participants volunteerly provided their codes, this is just for your information only.

https://www.dropbox.com/s/6ef461c96fcr0yr/knn3.py?dl=0

Background

Mobike, a smart bike sharing service introduced first to Beijing and then spread to many other cities in China, which has become increasingly popular among city residents and the best way of intra-urban commuting, in addition to public transit in many cities now. It has been credited with improving traffic conditions and reducing road congestion in cities. Nowadays, with more and more people are becoming environmentally concerned and green-minded, the population of Mobikers will only expand, thus to further realize the goal of making a big return of bicycles in cities. Mobike is devoted to facilitating short-distance travel with cutting-edge technology. Using machine learning to forecast travel destinations is one of its key applications to improve the efficiency of this bike share service.

So far more than 400,000 Mobike bicycles have been put in use in Beijing alone. A user can find available bicycles parked on the sidewalks, unlock one of them with his smartphone, ride to his destination, and then park and lock it again after he arriving. In order to better manage such a huge number of bicycles and deploy them to the places where they are most needed at any time, a precise prediction of the destination of any trip is needed.

Mission

Participants will use data sets provided by Mobike to predict destinations of Mobike trips.

The training set of data was collected from a given area in Beijing over a period of time in the past. Test set will be collected from the same area during a future period of time.

The test set will consist of a public set and a private set. During the competition, participants will only see the score they achieve with the public set, while their final standings on the leaderboard are determined by scores on the private set.

The labeled data set include three million pieces of data collected from over 300,000 Mobike users and 400,000 bicycles, including information such as the start location and time of a given trip, bicycle ID, bicycle type, and user ID. Participants are expected to predict which part of the city a user is heading to.