Geek Answers Handbook

Classification of multidimensional time series

I am currently working on attributes of 430 time series and approx. 80 thousand copies. Now I would like the binary class to classify each instance (not all ts). Everything I learned about TS classification spoke about labeling everything. Is it possible to classify each instance with some kind of SVM, completely ignoring the sequential nature of the data, or will this only lead to a really bad classifier? What other options exist that classify each instance, but still treat the data as time series?

+4
source share
2 answers

If the data is tagged, you may be lucky to combine the attributes together, so each instance will become one long time series and by applying the so-called Shapelet Transform . This will result in a vector of values ​​for each of the time series, which can be passed to SVM, Random Forest or any other classifier. It’s possible that choosing the right shapes will allow you to focus on one attribute when classifying instances.

If this is not checked, you can try the uncontrolled figures application to examine your data and continue the form conversion described above.

+2
source

, , 430 , , , .

, .. . , , f(t): Time -> Real. , 430 ( ) .

, (SVM, , , , ..) .

430 Maxent classifier ( ). N- , , . , " ".

, :

  • t-1, t-2...
  • ,
  • , , ... .

Maxent openNLP.

0

More articles:


All Articles