Simple algorythm for time series classification

Hi all,

I'm looking for a basic solution to cluster some time-series. I just begin using matlab, and am interested only simple algorithm, which compares the figure of 200 monthly observed groundwater data for 30 years and finally gives the class ID.

Thanks in advance

Zsolt

Answer 1

[Walter Roberson]

You could use k-means. A difficulty with k-means is that it requires that you input the number of classes it should cluster in to: if you do not know in advance how many classes, then you have a problem. I have heard that there are drivers that can run k-means several times to determine the "best" number of classes, and I have read that there are k-means-like techniques that figure the optimum out for themselves.

Do you know the number of classes ahead of time? Do you know the approximate characteristics of the classes ahead of time?

Answer 2

[Zsolt Zoltan Feher]

Well, k-means could be a good solution only if there isn't any relation between observations. My problem is that the groundwater shows autocorrelation, and can't handle the observations as individuals.