Hi there,

I am new to DM and more particularly to time series data mining and
I would like to get some help regarding a problem I am facing with
time series analysis.

given a set of n time series as training data, I want to be able to
construct a model that would represent some kind of average of the
given time series. This model would be used to conduct real time
detection of devaite input values.

e.g, if I have 10 time series each of which corresponds to samples
taken every 15 minutes during a period of 24 hours (360 samples for
each TS); I want to be able to detect every 15 minutes whether the
cuurent sample devaites from its corresponding value in the model
constructed from the training data.
And if the deviation is greater than some preconfigured treshold
(maybe %) value, I would like to be able to flag the deviation by
sending an alarm for example.

The challenge with this problem is that the detection of deviations
has to be conducted on a online (real-time) periodic basis , so I
cannot afford to wait till I gather all the 360 samples of today's
measurements before comparing them with the calculaed model.

I would apreciate if you guys can throw in some ideas to help me
tackle this challenging problems. your input would be really
appreciated.

Thanks.