Orange Forum • View topic - Identifying attributes that contribute most to predictions

Identifying attributes that contribute most to predictions

A place to ask questions about methods in Orange and how they are used and other general support.

Identifying attributes that contribute most to predictions

Postby soens » Fri Jul 11, 2014 19:54

Hello all,

I am using 22 attributes to train learners to predict a binary class state and I am wondering what is the optimal method to determine which of the 22 attributes are contributing most versus not contributing to the accurate prediction of the binary class.

Could anyone enlighten me if there is an automated method to have orange test every combination of my attributes to determine which combination is the best for predicting? Or do I have to do this manually selecting the attribute combinations one by one and comparing the AUCs?

Thank you,

Zach Soens
Grad Student
Baylor College of Medicine

Re: Identifying attributes that contribute most to predictio

Postby Ales » Tue Jul 15, 2014 13:37

soens wrote:I am using 22 attributes to train learners to predict a binary class state and I am wondering what is the optimal method to determine which of the 22 attributes are contributing most versus not contributing to the accurate prediction of the binary class.

There is no such general method. How much an attribute influences classification depends on the type of learning method.
soens wrote:Could anyone enlighten me if there is an automated method to have orange test every combination of my attributes to determine which combination is the best for predicting

No.


Return to Questions & Support



cron