If I run the logistic regression with X variables containing categorical data. (I do one-hot encoding on categorical data) How do I evaluate the variable importance? Is there any methods or literature on this?
Azen and Traxel's dominance analysis approach for logit regression could incorporate dummy or indicator coded categorical variables.
The indicator codes for each variable should be grouped together as a set in the analysis. That is, all the codes representing a single variable should always be included or excluded jointly.