Re: suitable inference for classification with fuzzy rule ensembles?

From: Carlos Andres Pena Reyes (penha@lslsun.epfl.ch)
Date: Fri Nov 09 2001 - 12:31:04 MET

  • Next message: Colette Faucher: "CFP for KES2002 Special Session : "Causality in Knowledge-Based Systems""

    Hi Chris,

    I also use EA to find (interpretable) fuzzy systems... I have no "answers to", but just opinions about your questions...

    "C. Setzkorn" wrote:
    > Lets say we predict a dichotomous attribute. Using Mamdami inference we
    :
    > Does the Mamdami inference mechanism actually make sense for the task of
    > classification?

    For such kind of well-established output classes I prefer to use singleton output values... given to the defuzzification stage, the output will still be continuous and a threshold can be applied to infer the final class.

    > Do the ?evolved? rules within a rule system actually mirror patterns
    > within a presented data set? (Since we treat the rule system more or
    > less like a black box.)

    Mmmmhh I'm not sure what do you mean with this question.

    > Does Mamdami inference presuppose that the domain of the attribute to be
    > controlled/predicted is actually continuous?

    Not necessarily, but in the case of discrete-valued input variables, the values should exhibit at least a conceptual monotonicity, in that way adjacent values belonging to a same fuzzy set will make sense (e.g., in the medical domain you can have the variable "familiar antecedents" with the values: None, second-familiar, first-familiar, contralateral, and homolateral; that satisfy the condition as they correspond to growing risk factors).

    > Are there more appropriated inference mechanisms for the task of
    > classification, especially when I try to predict categorical attributes
    > (not necessarily dichotomous)?

    Take a look of my paper Fuzzy Modeling by Fuzzy CoCo (see http://lslwww.epfl.ch/~penha/pub.html) where we present two different (but not new :-) approaches to the fuzzy classification of the Iris database, which has three classes.

    Please forward any answer you obtain as I'm interested in this subject.

    Best wishes,

    -- 
    Carlos Andrés PEÑA REYES
    Logic Systems Laboratory - LSL
    Computer Science Department - DI
    Swiss Federal Institute of Technology - EPFL
    CH 1015, Lausanne, Switzerland
    Email:	c.penha@ieee.org
    	penha@lslsun.epfl.ch
    http://lslwww.epfl.ch/~penha
    Tel: +41-21-693-2658
    Fax: +41-21-693 3705
    

    ############################################################################ This message was posted through the fuzzy mailing list. (1) To subscribe to this mailing list, send a message body of "SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at (2) To unsubscribe from this mailing list, send a message body of "UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com" to listproc@dbai.tuwien.ac.at (3) To reach the human who maintains the list, send mail to fuzzy-owner@dbai.tuwien.ac.at (4) WWW access and other information on Fuzzy Sets and Logic see http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info (5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html



    This archive was generated by hypermail 2b30 : Fri Nov 09 2001 - 12:50:43 MET