Dear Team,
I am going through a series of videos where WOE/IV (weight of evidence & information value) has been used to highlight outliers and missing values. These techniques are useful for both nominal and continuous variables.
There is another techniques contended only for continuous variables i.e. VIF (variation inflation factor) so i assistance as to is there a thumb rule during the pre phase of a model building process which is more robust and what to use when.
Please advice.
and i did miss to add WOE/ IV are very important variable reduction techniques. But my question is from a correlation point in IV's
Dear Bhatia,
WOE/IV are normally used for binary outcomes and based on my personal experience, I have never come across WOE/IV being used as variable reduction tecniques in predicting continuous variables. VIF may be an indirect variable reduction technique whose main function is to indicate the multicollinearity present in your IVs used for predictive modelling. The correlation matrix could be the best pre phase strategy to understand the degree of correlation among IVs and select appropriate IVs based on suitable cutoff R-value.
HI Anush, totally in sync however we can also use WOE/IV for continuous variable where it creates bins for these variables. Please see link below:
http://multithreaded.stitchfix.com/blog/2015/08/13/weight-of-evidence/
As far as I'm concerned,WOE is a transformation to make predictor variable linear.IV is used to select certain variables which go into the logistic model.VIF is used to detect the multicollinearity present in your variables.
- Badges
- Science topic