====== Features ======
===== Statistik =====
Für jede Dimension (Aggregate):
  * Mittelwert 
  * Standardabweichung
  * Skewness, Kurtosis, höhere Momente
  * Maxima, Minima

Zeitreihensicht:
  * Crosscorrelation zwischen Dimensionen und Autokorrelation
  * Ordnung AR/MA, Parameter

Modellparameter

http://stats.stackexchange.com/questions/50807/features-for-time-series-classification
http://nips.cc/Conferences/2012/Program/event.php?ID=3217
http://robjhyndman.com/hyndsight/tscharacteristics/
====== Transformationen ======
===== SAX =====
http://stats.stackexchange.com/questions/3238/time-series-clustering-in-r
===== SVD =====
  * SVD: Problem bei verschieden langen Zeitreihen
  * Vorschläge: 
    * Auffüllen durch resamplen vergangener Werte
    * http://videolectures.net/mlss09us_candes_mccota/

http://stats.stackexchange.com/questions/1268/svd-dimensionality-reduction-for-time-series-of-different-length?rq=1

===== DFT =====
http://stats.stackexchange.com/questions/51475/series-dimensionality-reduction-for-classification-input?rq=1
http://stats.stackexchange.com/questions/9475/time-series-clustering/19042#19042
https://www.youtube.com/watch?v=wd6hAPS4ILc
===== Wavelet =====
http://stats.stackexchange.com/questions/51475/series-dimensionality-reduction-for-classification-input?rq=1
http://stats.stackexchange.com/questions/39007/clustering-time-series-with-wavelets-in-r?rq=1

===== Distanzmaße =====

==== Dynamic Time Warping ====
KNN + Dynamic Time Warping. 
http://stats.stackexchange.com/questions/66027/time-series-classification-very-poor-results?rq=1