Being given the known criteria (a training set composed of the classified samples), you can try to find the appropriate curve using evolutionary strategy / genetic algorithm (I'd rather choose real-coded genetic algorithms over binary one).

Hello Omar,

To develop a fuzzy classification system, the most important task is to construct membership functions and to find a set of suitable fuzzy rules in the fuzzy classification system. There are two approaches to obtain fuzzy rules. One of them is given directly by domain experts; the other is obtained through a machine learning process based on training instances. In recent years, many methods have been proposed to generate fuzzy rules from training instances.

Please go through herewith the attached file which i have performed in Matlab

But what is the way you use fuzzy sets to feature selection? More information is needed.

Good point Grzegorz. I guess this is more classifier to classify features like Suresh Merugu wrote above than "feature selection", but in any case, automated membership extraction should be doable using GA/RCGA/EA/MA. Let's hear the author's explanation first then!

But, i mean fuzzy sets of feature selection,if i have data set without predefined rules

I don't get your point - do you have a feature dataset? If so, the problem is to "extract correct rules" to cover the data set, isn't it? If so, you can try one of the techniques of the automated rule generation. It strongly depends on the Fuzzy Inference System you choose. The common part is to invent such rules that from one point of view they cover the part of the training dataset and from the other are not so strict (i.e. covering only one sample) to avoid overlearning.

Thanks a lot, this file include an overview in membership function generation