Clustering Washington D.C. bike share stations by using time series data

Abstract

In this economic era, technological developments lead on changing consumption habits. This mutation in consumption created sharing economy, which has consistently been getting a bigger role in economy day by day. Bike sharing systems, as one of the pioneers of sharing economy, are now counted as type of transportation in most of the big cities. Thus there are various studies works on improving efficiency of these systems. This thesis aims to cluster Washington D.C. bike-share stations and by identifying stations with similar rental behaviours, it is aimed to impact on efficiency of management of these stations. The dataset includes one-year rental records of shared bikes with their date time information as large time series data. Before clustering bike stations, first part of the thesis describes various clustering methods. Clustering is unsupervised learning method, to organize set of observations by their similarity and classify them. Calculating similarity between observations depends on selected clustering approach. This thesis covers iteration steps and evaluation of several clustering approaches, like k-means, k-medoids, hierarchical clustering, model based clustering, DBSCAN. After exploration of common algorithms, in the second part, to cluster bike-share stations, k-means, agglomerative hierarchical clustering and model based clustering methods implemented and evaluated.Αυτή η διατριβή στοχεύει να συγκεντρώσει σταθμούς κοινοποίησης ποδηλάτων στην Ουάσιγκτον και προσδιορίζοντας σταθμούς με παρόμοιες συμπεριφορές ενοικίασης, έχει ως στόχο να επηρεάσει την αποτελεσματικότητα της διαχείρισης αυτών των σταθμών.