KIT | KIT-Bibliothek | Impressum | Datenschutz

Weighted aggregation in the domain of crowd-based road condition monitoring

Laubis, Kevin; Simko, Viliam; Weinhardt, Christof


This paper focuses on crowd-based road condition monitoring using smart devices, such as smartphones and evaluates different strategies for aggregating multiple measurements (arithmetic mean and weighted means using R2 and RMSE) for predicting the longitudinal road roughness. The results confirm that aggregating predictions from single drives leads to a higher model performance. This has been expected and confirms the intuition. The overall R2 could be increased from 0.69 to 0.75 on average and the NRMSE could be decreased from 9% to 8% on average. However, contrary to the intuition, the results show that weighted aggregations of single predictions should be avoided, which is consistent with previous findings in other domains, such as financial forecasting.

Volltext §
DOI: 10.5445/IR/1000061981
Veröffentlicht am 01.11.2017
Cover der Publikation
Zugehörige Institution(en) am KIT Universität Karlsruhe (TH) – Einrichtungen in Verbindung mit der Universität (Einrichtungen in Verbindung mit der Universität)
FZI Forschungszentrum Informatik (FZI)
Fakultät für Wirtschaftswissenschaften – Institut für Informationswirtschaft und Marketing (IISM)
Publikationstyp Proceedingsbeitrag
Publikationsjahr 2016
Sprache Englisch
Identifikator ISBN: 978-3-88579-653-4
ISSN: 1617-5468
KITopen-ID: 1000061981
Erschienen in INFORMATIK 2016 : Informatik von Menschen für Menschen, 46. Jahrestagung der Gesellschaft für Informatik, 26.-30. September 2016, Klagenfurt. Hrsg.: H. C. Mayr
Verlag Gesellschaft für Informatik (GI)
Seiten 385-393
Serie GI-Edition / Proceedings. Lecture Notes in Informatics ; 259
Externe Relationen Abstract/Volltext
Schlagwörter Crowd-based sensing, road condition monitoring, international roughness index, predictive road maintenance, weighted aggregation, ensemble learning
KIT – Die Forschungsuniversität in der Helmholtz-Gemeinschaft
KITopen Landing Page