The Altair Community is migrating to a new platform to provide a better experience for you. In preparation for the migration, the Altair Community is on read-only mode from October 28 - November 6, 2024. Technical support via cases will continue to work as is. For any urgent requests from Students/Faculty members, please submit the form linked here

SVM Multiclass Classification

kaalgotakaalgota Member Posts: 6 Contributor II
edited June 2019 in Help
How SVM will create hyperplane for multiclass label?
Tagged:

Answers

  • MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    It can't, you have to use other techniques, e.g. the 1-vs.-all pattern as performed by the Polynominal by Binominal regression.
    Implementations of the SVM that support multiple classes do something similar internally.


    Best regards,
    Marius
  • mbeckmannmbeckmann Member Posts: 15 Contributor II

    Unfortunatelly I have to disagree,

     

    LIBSVM is able to deal with multiclass classification, and if fact there's a connector to LIBSVM in RM, but then the LIBSVM results stays constrained by this strange architecture of Polynomial and Binomail that exists only in Rapidminer.

     

    Even a simple IRIS problem, is not possible to obtain a simple AUC, F-Measure, or ROC, and the Polynomial by Binomial classification or regression, simply doesn't work.

     

    There is mathematical formulation to abandon this polynomial and binomial architecture, and this is a widespread methodology in any machine learning tool like Weka, R, Phyton libs, etc.

    I think it's time to abandon this Poly/Bin architecture, but this involves deep changes in the RM structure.

     

     

     

     

  • mbeckmannmbeckmann Member Posts: 15 Contributor II

    Hi,

     

    The Polynomial by Binomial classification or regression, is working fine, and I can't disagree and complain about Poly/Bin architecture, without to propose possible solutions and to contribute some how with this great platform. Possible solutions for this problem are discussed in this topic.

     

     

     

     

     

Sign In or Register to comment.