Classification of the vélib stations using Kmeans, Dynamic Time Wraping and DBA averaging method

The Bike Sharing System (BSS) has become a more and more popular means of transport in Paris and in many other cities around the world. It is also generating an increasingly huge amount of data describing users trips. Such datasets may be very useful for the data mining community in order to improve the global performance of the BSS. In this paper, we focus on the resources availability (free docks and available bikes) in the Parisian BSS called Velib´. We analyze a Velib´ trip dataset in order to separate the Velib´ stations into three categories (balanced, overloaded and underloaded clusters), according to the ratio between arrivals and departures relative to each station, during the whole day. For this purpose, we use the well known Kmeans clustering algorithm, along with the Dynamic Time Wraping (DTW) metric to measure the similarity between the clusters. We choose to update the centers of the clusters using the efficient Dtw Barycenter Averaging (DBA) method.