Statistical and Numerical Algorithms for Time Series Classification
Cluster analysis of time series is usually performed by extracting and comparing relevant interesting features from the data. Quite a few numerical algorithms are available that search for highly separated data sets with strong internal cohesion so that some suitable objective function is minimized or maximized. Algorithms developed for classifying independent sample data are often adapted to the time series framework. On the other hand time series are dependent data and their statistical properties may serve to drive allocation of time series to groups by performing formal statistical tests. Which is the class of methods that has better chance of fulfilling its task is an open question. Some comparisons are presented concerned with the application of different algorithms to simulated time series. The data recorded for monitoring the visitors flow to archaeological areas, museums, and other sites of interest for displaying Italy’s cultural heritage are examined in some details.
KeywordsTime Series Cross Correlation Visitor Flow Autoregressive Moving Average Time Series Classification
This work was financially supported by grants from MIUR, Italy. R. Baragona gratefully acknowledges financial support from the EU Commission through MRTN-CT-2006-034270 COMISEF.
- Gómez, V., & Maravall, A. (1996). Programs TRAMO and SEATS: instructions for users, Technical Report 9628, The Banco de España, Servicios de Estudios.Google Scholar
- Piccolo, D. (1989). On the measure of dissimilarity between ARIMA models. In Proceedings of the A. S. A. Meetings, Business and Economic Statistics Sect., Washington D. C., 231–236.Google Scholar
- Sarno, E. (2001). Further results on the asymptotic distribution of the Euclidean distance between MA models. Quaderni di Statistica,3, 165–175.Google Scholar
- Zani, S. (1983). Osservazioni sulle serie storiche multiple e l’analisi dei gruppi. In D. Piccolo (Ed.), Analisi Moderna delle Serie Storiche (pp. 263–274). Milano: Franco Angeli.Google Scholar