0
$\begingroup$

I am new to statistics. I have some programming background but am a novice in statistics. I need some help/direction identifying methods/technics to solve below problem.

I have multiple time series, around 100. Each time series potentially have 100K to 1M readings/measurements at equal time intervals. Out of which, all time series are approximately similar (please note approximately similar not necessarily exactly similar) except one which will exhibit approximately similar pattern as others until some time and then starts to deviate. Like shown in the picture below.

In the picture below 1 and 2 are approximately similar time series. Because of large number of readings on a smaller scale, the small differences are not noticeable or not visible. But from my study point of view, I need to find 1 and 2 identical. And notice time series 3, it exhibits same pattern until the time where it is shown in red oval and from there it starts to deviate.

enter image description here

My question is what methods/technics I can use to identify the one time series that is deviating and the point/reading from where it starts to deviate?

Thanks in advance for any help.

Srinivas

$\endgroup$
  • $\begingroup$ Is cointegration a possible solution for this problem? Need some expert advice. $\endgroup$ – Srinivas Feb 17 '17 at 6:58
  • $\begingroup$ After further analysis, it doesn't look like a time series problem. This series has no Trend and Seasonality that are required of a time series to apply time series methods. If not time series methods, are there any other methods/techniques that I can apply? $\endgroup$ – Srinivas Mar 23 '17 at 8:36
0
$\begingroup$

One solution is to statistically determine points at which changes occur in the short term mean or variance along each timeseries. Change points from each of the 100 timeseries should occur at similar locations until a timeseries starts to deviate. The R changepoint package implements a number of change point detection methods you could apply to this task.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.