Determining significance of predictors in R

datascience
statistics

#1

Hello all,
I would like to know the importance and relative importance of predictors(say 100 in number) without having to build a model like random forest.
Is it possible ? If yes, how ?

Thanks.


#2

Hi NSS - Azure Machine Learning has a module called ‘Filter Based Feature Selection’ that helps in selecting features with the highest predictive power. Pls. check out this link: https://msdn.microsoft.com/en-us/library/azure/dn913071.aspx. Even if you are not using Azure ML, the information provided on the web page will be useful to devise a way of finding relative feature importance. Hope this helps.

–Karthik


#3

Just to complete my previous post - one can also use packages like the Boruta package in R for feature selection - http://www.analyticsvidhya.com/blog/2016/03/select-important-variables-boruta-package/. Having said that Boruta is just a wrapper on top of Random Forests algorithm.


#4

@skkeyan

Thank you for the answer. It helped :slight_smile:


#5

Hello,

You can try something different with package “woe”…

https://cloud.r-project.org/web/packages/woe/index.html

Kind Regards,
Carlos Ortega


#6

You can try information value .