0
$\begingroup$

I have a data.frame which has multiple time series in it, in the following manner:

01.01.16 Value1-Time-Series-1 Value2-Time-Series-1 Value3-Time-Series-1 ID-1
01.02.16 Value1-Time-Series-1 Value2-Time-Series-1 Value3-Time-Series-1 ID-1
01.03.16 Value1-Time-Series-1 Value2-Time-Series-1 Value3-Time-Series-1 ID-1
01.01.16 Value1-Time-Series-2 Value2-Time-Series-2 Value3-Time-Series-2 ID-2
01.02.16 Value1-Time-Series-2 Value2-Time-Series-2 Value3-Time-Series-2 ID-2
01.03.16 Value1-Time-Series-2 Value2-Time-Series-2 Value3-Time-Series-2 ID-2

Meaning:

  • There is one timeframe which is the same for every time series
  • There are multiple (thousands) of time series within that time
  • All are made up by the same type of values (columns)

I want to cluster this dataset in order to find the time series objects similar to each other. I want to find those time series(s) which changed the same with respect to the values mentioned. I'm obviously interested in the measured distance. But using a simple hierarchical clustering algorithm is not cutting it since I need to tell the (whatever clustering algorithm I end up using) what is a time series and what not.

Any ideas how to this?

I'm using R for this. If there might be a better option with Python, I'm totally open to use it.

Thanks for any kind of help!

$\endgroup$
3
$\begingroup$

For an applied solution to your problem, I highly recommend reading the following: TSclust: An R Package for Time Series Clustering which can be found here. It presents the TSclust package in R and provides code.

Note that time series data is special, and cannot be treated like other data. So typical clustering techniques are not appropriate.

|improve this answer|||||
$\endgroup$

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.