Multicollinearity technique for any regression model



Do we need to first build the model. Either It’s binary logistic regression or random forest and then run VIC, PCA ridge etc on the model. Is checking and resolving technique same for every model?


Hi @jayanthd,

You can first check the correlation between variables and try to find the variables having more correlation. This will tell you whether there is multicollinearity in your model. If you find any multicollinearity, you can remove the most correlated variables, i.e. if variable A and B have high correlation you can remove any one of the variable and make the model.

This may vary from model to model. You can first build the model on all set of variables and then try to find multicollinearity or you can first check the multicollinearity and after removing the most correlated variables, make your model. It totally depends on how you explore your data.


What is VIC?


My Apologies. Its typo its VIF Variance inflation factor.