I am interested in forming features to do predictive analysis based on small data. The primary features that I have are numeric variables and a few of them are categorical features.
I am sorry, still in lack of information ... The paper cited below describes an approach of hierarchical conceptual clustering and using it principal component analysis for mixed feature data. Hope you will find it useful.
Ichino, M., & Yaguchi, H. (1994). Generalized Minkowski metrics for mixed feature-type data analysis. IEEE Transactions on Systems, Man, and Cybernetics, 24(4), 698-708.