Hi Wassila, 

Several evaluation methods that are similar to the "SVMAttributEval (that you mentioned) which are:

- CfsSubsetEval

-ChiSquaredAttributeEval

-ClassifierSubsetEval

-ConsistencySubsetEval

-CostSensitiveAttributeEval

-CostSensitiveSubsetEval

-FilteredAttributeEval

-FilteredSubsetEval

-GainRatioAttributeEval

-InfoGainAttributeEval

-LatentSemanticAnalysis

-OneRAttributeEval

-PrincipalComponents

-ReliefFAttributeEval

-SymmetricalUncertAttributeEval

-WrapperSubsetEval

The package called "classifierBasedAttributeSelection" is available here:

http://prdownloads.sourceforge.net/weka/classifierBasedAttributeSelection1.0.5.zip?download

HTH.

Samer

thank you samer sarsam, i want to know about embedded feature selection algorithms. i know some of the above listed methods like infogain, relief, they are filter.

for "classifierBasedAttributeSelection" you mean this is an embedded method in which we just select the underlying classifier?

Thanks.

You're welcome, Wassila.

classifierBasedAttributeSelection is a package that contains all the required evaluation methods to be used during attribute selection process, get stored in a zip file format. Those methods are similar to "SVMAttributEval" that you know.

HTH.

Samer

Thanks, i am looking for a specific type of feature selection algorithms which are the embedded ones. i know one of them SVMAttributEval.

please can any one give me some others embedded FS methods in weka.

Feature selection in WEKA is a combination of "search method" (e.g, GreedyStepwise) and "evaluation method" (e.g., SVMAttributEval). Therefore, to be precise, all the highlighted methods, so far, are evaluators to the subset features being selected by the search method. Finally, once evaluating the merit of the selected subset, classification algorithm builds prediction model using those feature. 

HTH.

Samer

yes i agree with you, all  feature slection algorithms include these two steps in their process : search and evaluation steps. based on how to evaluate the candidate feature subset during the search process, FS algorithms are classified into : filters wrappers and embedded.

Embedded methods (are like wrapper but they are algorithms that build a model and when building that model they select the most relevant features).

SVMAttributEval is an embedded FS algorithm (proposed in:  I. Guyon, J. Weston, S. Barnhill, V. Vapnik (2002). Gene selection for cancer classification using support vector machines. Machine Learning. 46:389-422.) in weka, we can select this in the "evalaution" drop list and in the search list we select a ranker method.

Another embedded Fs method which is : FS-P (Feature Selection—Perceptron), proposed in  : Mejía-Lavalle M, Sucar E, ArroyoG(2006) Feature selectionwith a perceptron neural net. In: Proceedings of the international workshop on feature selection for data mining, pp 131–135

but i didn't find it in weka ????

In WEKA, the most similar one is "SVMAttributeEval". Knowing that, feature selection using Perceptron is not available so far. My advice is to use "AttributeSelectedClassifier" in the "Classify" panel where you can apply "MultilayerPerceptron", "SVMAttributeEval", and "Ranker" together.

HTH.

Samer

I will try this, thank you for your help.