Validating the coverage of bus schedules: A Machine Learning approach

• Type: Article in International Scientific Journal
• Date: 2015

Authors

Joao Mendes Moreira (Author) FEUP; Luis Moreira Matias (Author) Other; Joao Gama (Author) FEP; Jorge Freire de Sousa (Author) FEUP

Journal

Title: Information Sciences

Vol. 293 No. 1

Pages: 299-313

ISSN: 0020-0255

Publisher: Elsevier

Indexing

ISI Web of Knowledge - 34 Citations

Scopus - 37 Citations

Scientific classification

FOS: Natural sciences > Computer and information sciences

Other information

Authenticus ID: P-00A-0V5

DOI: 10.1016/j.ins.2014.09.005

Abstract (EN): Nowadays, every public transportation company uses Automatic Vehicle Location (AVL) systems to track the services provided by each vehicle. Such information can be used to improve operational planning. This paper describes an AVL-based evaluation framework to test whether the actual Schedule Plan fits, in terms of days covered by each schedule, the network's operational conditions. Firstly, clustering is employed to group days with similar profiles in terms of travel times (this is done for each different route). Secondly, consensus clustering is used to obtain a unique set of clusters for all routes. Finally, a set of rules about the groups content is drawn based on appropriate decision variables. Each group will correspond to a different schedule and the rules identify the days covered by each schedule. This methodology is simultaneously an evaluator of the schedules that are offered by the company (regarding its coverage) and an advisor on possible changes to such offer. It was tested by using data collected for one year in a company running in Porto, Portugal. The results are sound. The main contribution of this paper is that it proposes a way to combine Machine Learning techniques to add a novel dimension to the Schedule Plan evaluation methods: the day coverage. Such approach meets no parallel in the current literature.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 15