Theory on methods of selection of relevant variables

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

Theory on methods of selection of relevant variables

Juan Sebastian Mejia
Hi guys, in my project I used the following methods to select relevant variables:

1) Attribute evaluator: CFsSubsetEval with 10 folds BestFirst and Cross-Validation search method.

2) Attribute evaluator: ReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method.

3) Attribute evaluator: fReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method, with NumericToNominal filter.

4) Classifier: ReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method, with NumericToNominal filter.

I want to know where I can find theory or literature regarding each and every one of these methods, since the project evaluators require me to attach this information in the theoretical framework and I need to reference all the theory that I find. I hope you can help me guys, thanks in advance.

_______________________________________________
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: Theory on methods of selection of relevant variables

Eibe Frank-2
Administrator
From the information available under “More…” in the GUI:

CfsSubsetEval:

M. A. Hall (1998). Correlation-based Feature Subset Selection for Machine Learning. Hamilton, New Zealand.

ReliefFAttributeEval:

Kenji Kira, Larry A. Rendell: A Practical Approach to Feature Selection. In: Ninth International Workshop on Machine Learning, 249-256, 1992.

Igor Kononenko: Estimating Attributes: Analysis and Extensions of RELIEF. In: European Conference on Machine Learning, 171-182, 1994.

There is no “theory” paper for the cross-validation in the attribute selection panel (if that’s what you mean). With this “attribute selection mode”, the attribute selection method is just run on the k different training folds that occur in a k-fold cross-validation (the test folds are not used). This will give you an idea of the robustness of the selection process with respect to changes in the training data. In the case of a subset evaluator, you will see for how many of the k folds a particular attribute has been selected. In the case of an attribute evaluator, you will see the average rank and the average “merit” of each attribute (plus the corresponding standard deviations).

The Ranker method simply ranks the attributes based on the merit provided by the selected attribute evaluator. You can either select a certain number of top-N attributes or you can cut off by providing a threshold on the attributes’ merit.

Cheers,
Eibe

> On 28/08/2019, at 1:23 PM, Juan Sebastian Mejia <[hidden email]> wrote:
>
> Hi guys, in my project I used the following methods to select relevant variables:
>
> 1) Attribute evaluator: CFsSubsetEval with 10 folds BestFirst and Cross-Validation search method.
>
> 2) Attribute evaluator: ReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method.
>
> 3) Attribute evaluator: fReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method, with NumericToNominal filter.
>
> 4) Classifier: ReliefFAttributeEval with 10 folds Ranker and Cross-Validation search method, with NumericToNominal filter.
>
> I want to know where I can find theory or literature regarding each and every one of these methods, since the project evaluators require me to attach this information in the theoretical framework and I need to reference all the theory that I find. I hope you can help me guys, thanks in advance.
> _______________________________________________
> 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