I have a time series classification task to solve, based on statistical features like maximum, minimum, standard deviation, etc. It is a multiclass classification and my estimator seems to be confused with the two classes shown in picture. It can more or less well identify all the other classes, but fails on these ones.

Do you have any suggestions which statistical features would be suitable to make the two types of time series shown in the picture seperatable? I am not too much into statistics and am sure, there is something I just don't know. What my model already includes is:

mean, median, minimum, maximum, standard deviation, number of values below 50.

I am happy to get some inspiration about features I could extract from the time series in order they improve my model!

EDIT: The values of the time series differ quite a lot. Some have ranges over 70000, others between 0 and 100 (2nd picture)

I normalized the time series of all classes to visualize them in one plot (3rd picture)

Example of time series to be seperated Example of all types of time series Example of normalized timeseries

  • $\begingroup$ This seems impossible to answer without having the context of the other time series in your data, for otherwise just about any characteristic would serve to distinguish these two. $\endgroup$ – whuber Aug 14 at 15:42
  • $\begingroup$ Thanks for the hint. I added two more pictures to show the other classes. 1st picture: Time series that i need more features for 2nd picture: All classes with real scale 3rd picture: all classes normalized $\endgroup$ – nopact Aug 14 at 15:51

Your Answer

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

Browse other questions tagged or ask your own question.