I am having a mixed-Numeric and Categorical variable, can I use variable clustering algorithms like k-means or some other algo. for the variable reduction. If it is possible then how will I select the variable for further analysis? For Example, if I have 20 features in my data set (mixed type ) how can I use only those feature which will highly influence my logistic regression model.
Your problem comes under Dimensionality Reduction - see if any approach from below article can help: