Hitachi Vantara Pentaho Community Forums
Results 1 to 5 of 5

Thread: Trying to find a multiple classifier for known and unknown category outcomes

  1. #1
    Join Date
    Mar 2014
    Posts
    5

    Default Trying to find a multiple classifier for known and unknown category outcomes

    Hi,

    I'm a biologist trying to develop a capture-mark-recapture survey method using animal vocalisations. I've only just started using WEKA. I've formerly used Statistica to run artificial neural networks - MLPs - to show that I can classify vocalisations to individuals with up to 100% accuracy. But if I have new individuals present in the validation database that weren't in the training database, it just classifies them to the nearest individual in the training database. I need it to go to an "unknown" category and have read Probabilistic Neural Networks can do this.

    So I have the variables, the correct identity and the knowledge that I can classify them correctly with up to 100% accuracy. What I want is a dual-layer outcome:

    1) Identity: known or unknown
    and then if identity is known
    2) Identity = individual A (confidence level = X)

    Can anyone let me know how this would work in WEKA? As I say I'm a novice so as much detail as possible would be wonderful!

    Thanks!

  2. #2
    Join Date
    Aug 2006
    Posts
    1,741

    Default

    Assuming that unknown individuals are reasonably different from the known ones in the training set, I guess you could try applying an anomaly detection method in step 1. There are several one class learning methods in Weka that can be applied to outlier/anomaly detection problems:

    LOF (Local Outlier Factor)
    http://weka.sourceforge.net/packageM...or/Latest.html

    OneClassClassifier:
    http://weka.sourceforge.net/packageM...er/Latest.html

    Cheers,
    Mark.

  3. #3
    Join Date
    Mar 2014
    Posts
    5

    Default

    Thanks! I'll try that.

    Just a quick follow up - are naive bayes classifiers the same thing as probabilistic neural networks?

    Thanks again,

    Holly

  4. #4
    Join Date
    Aug 2006
    Posts
    1,741

    Default

    No, naive Bayes classifiers can be thought of as a very limited bayesian network, where each attribute has just one parent in the graph - the class attribute.

    Cheers,
    Mark.

  5. #5
    Join Date
    Mar 2014
    Posts
    5

    Default

    Thanks the internet has lied to me then.

    Last question (for now!): Does WEKA have Probabilistic Neural Networks? One of the papers I've read used them and it'd be interesting to compare to my results.

    Thanks,

    Holly

Tags for this Thread

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
Privacy Policy | Legal Notices | Safe Harbor Privacy Policy

Copyright © 2005 - 2019 Hitachi Vantara Corporation. All Rights Reserved.