Hi All,
I have a question regarding clustering techniques and whether it is
appropriate for my problem.
I have a number, say 30, overlapping time series response curves for a
month's worth of data. If I plot them individually it is clear that the
first 10 or so exhibit a strong weekly cyclical pattern, in other words they
are very similar. The remaining time series curves from 10 upwards exhibit
the same cyclical pattern but it gets gradually weeker until you get to the
30th curve which just looks like noise. I would like to combine the 10 or so
"similar" curves together as my baseline.
My question is what should I use as my cut-off point?
A visual inspection seems somewhat subjective. Can I use a standard
clustering method to group the "similar" time series curves?
Regards,
Richard
|