Ordinal time series analysis

(C.Bandt, Greifswald)
 

In a study of EEGs of children's hospital at Greifswald University, Kendall's tau was tested as an autocorrelation and crosscorrelation function. It turned out that it is robust with respect to most types of artefacts, and can be computed much faster than ordinary correlation and FFT. Our version of tau easily generalizes to higher-dimensional correlations. In the context of dynamical systems this leads to interesting theoretical problems. Another ordinal quantity has the properties of a frequency spectrum.

Since these parameters seem to be unknown in time series analysis, they should be studied from a theoretical viewpoint and tested at various types of data. We shall obtain at least some fast methods for data mining of huge time series collections, as well as for real-time analysis of multi-channel-EEG. Moreover, a theory of ordinal time series ought to be established.

With B. Pompe und K. Keller we have a tight cooperation in Greifswald. It is planned to cooperate also with the groups in Freiburg, Ilmenau, Jena, Kiel und Potsdam who also work with EEG data.