Concept Drift

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Concept Drift

Chaitanya Birudavolu
Hi,

Is there any support for Concept Drift in Weka (or in the packages of Weka)?

--1--Any ways of generating data with simulated concept drift ?
--2--Any classifiers or meta-classifiers or wrappers that can handle concept drift?
--3--Any ways to evaluate and visualize performance in the presence of concept drift?

I reckon that in the real world, drifting concepts and hidden contexts must be far more common than the neat text-book scenario of stationary concepts.

So, is there anything that I can do in (or with) Weka for Concept Drift?

I'm aware of the following, but I don't think they fully address the concept drift scenario (please correct me if I'm wrong there):

-Updateable Classifiers in Weka
-MOA (to some extent, from the little coverage in the course "Advanced Data Mining with Weka")

Thanks and Regards,
Chaitanya Birudavolu
_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
Reply | Threaded
Open this post in threaded view
|

Re: Concept Drift

Mark Hall
Weka doesn't have any support for detecting or dealing with concept drift. Take a look at the MOA project:

https://moa.cms.waikato.ac.nz/

This does have schemes that handle concept drift.

I think that common practice for dealing with concept drift in real deployments is simply to monitor predictive performance and periodically re-build models on recent historical data when performance drops below some acceptable level.

Cheers,
Mark.

´╗┐On 29/11/18, 4:52 AM, "Chaitanya Birudavolu" <[hidden email] on behalf of [hidden email]> wrote:

    Hi,
   
    Is there any support for Concept Drift in Weka (or in the packages of Weka)?
   
    --1--Any ways of generating data with simulated concept drift ?
    --2--Any classifiers or meta-classifiers or wrappers that can handle concept drift?
    --3--Any ways to evaluate and visualize performance in the presence of concept drift?
   
    I reckon that in the real world, drifting concepts and hidden contexts must be far more common than the neat text-book scenario of stationary concepts.
   
    So, is there anything that I can do in (or with) Weka for Concept Drift?
   
    I'm aware of the following, but I don't think they fully address the concept drift scenario (please correct me if I'm wrong there):
   
    -Updateable Classifiers in Weka
    -MOA (to some extent, from the little coverage in the course "Advanced Data Mining with Weka")
   
    Thanks and Regards,
    Chaitanya Birudavolu
    _______________________________________________
    Wekalist mailing list
    Send posts to: [hidden email]
    To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
    List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
   


_______________________________________________
Wekalist mailing list
Send posts to: [hidden email]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html