Article citationsMore >>

Hyndman, R., Khandakar Y., Automatic time series forecasting: The forecast package for r. Journal of Statistical Software, (26(3)), 2008.

has been cited by the following article:

Article

Application of Spatio-Temporal Clustering in Forecasting Optimization of Geo-Referenced Time Series

1Dipartimento di Informatica, Università degli Studi di Bari Aldo Moro via Orabona, Bari, Italy

2Faculty of Information Technology, Mediterranean University, Vaka Djurovica b.b. Podgorica, Montenegro


American Journal of Modeling and Optimization. 2014, Vol. 2 No. 1, 8-15
DOI: 10.12691/ajmo-2-1-2
Copyright © 2014 Science and Education Publishing

Cite this paper:
Sonja Pravilovic, Annalisa Appice. Application of Spatio-Temporal Clustering in Forecasting Optimization of Geo-Referenced Time Series. American Journal of Modeling and Optimization. 2014; 2(1):8-15. doi: 10.12691/ajmo-2-1-2.

Correspondence to: Sonja  Pravilovic, Dipartimento di Informatica, Università degli Studi di Bari Aldo Moro via Orabona, Bari, Italy. Email: sonja.pravilovic@uniba.it

Abstract

A novel field of data mining has been spatio-temporal clustering focused on the new methods and techniques, which are able to adapt previous methods and solutions to the new problems. A set of geo-referenced time series are data generated by several devices like GPS, sensor station, cell phones and many other sensing device. This paper defines the the new K-means clustering grouping spatially and temporally correlated geo-referenced time series obtained from sensors in a specific geographic area. For all time series in the cluster, choosing the best forecasting parameters, we apply one of the most accurate and most efficient forecasting models of time series called ARIMA. This paper investigates a new mechanism to determine spatio-temporal distances measure between sensor stations in the same spatio-temporal neighborhood (cluster). By calculating, the best forecasting parameters applied for all time series in the same cluster proposed algorithm obtains more accurate and more efficient forecasting results, than forecasting time series independently one from other in space and time. We studied the accuracy of proposed model comparing it to the already known applied to compute prediction of time series and applying it to real life data.

Keywords