🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉

Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.

CLICK HERE TO DOWNLOAD

Precision/Recall when using multiclass svm

TimLTimL Member Posts: 2 Contributor I
edited November 2018 in Help
Hey guys,

For a concept exctraction process task I am using a multiclass svm classifier. In most of the paper I have read, they evaluate their results by using precision/recall/f-measure etc.

Now ofcourse I use the 'polynomial by binomial classification' block, to make the svm do multiple classes. For performance I tried the 'Binominal Classification Performance' block however I get two  problems.
1) If I use it outside the 'polynomial by binomial classification' block, it says that the label is not binominal and thus it cant measure the performance.
2) If I use it inside the 'polynomial by binomial classification' block, it just doesnt show any performance at all.

I am almost certain that it is possible to do this, since I have read it different studies. Does anyone have an idea?

Cheers!

Answers

  • mschmitzmschmitz Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,085  RM Data Scientist
    Hi TimL,

    why do you want to use a binominal performance for a polynominal problem? I would suggest using the Performance (Classification) operator for this task. Of course this needs to be done on the testing side of your validation.

    Cheers,
    Martin

    - Head of Data Science Services at RapidMiner -
    Dortmund, Germany
  • TimLTimL Member Posts: 2 Contributor I
    Hey Martin,

    Thanks for your responds!

    The values that I am working with are strings, and SVM cant handle polynominal values. However I just can not seem to get it to work. My input is basically five columns with words in it, one column being the classification. I have been trying every single combination but I cant get it to work.

    Any suggestions?

    Tim
Sign In or Register to comment.