hello friend

first you extract all the features.

second, you start with null features.

third, keep adding the features to the null set based on the information gain.

if the gain is reduced, then discard the feature else keep it. and the process continues till all the variables are exhausted.

I hope this helps you. I have used this in my papers.

Best regards

Dr.Indrajit Mandal

Dear Indrajit

With this approach the execution take a long time because it is an exhaustive research. there is a function in matlab who do the forward feature selection  (The link that @Hana Shaker share) but i did understand how to project it in biometrics because i must select the most relevant feature for all person in the database.

could you sent me your paper please?

Best regards

Lamis