Depending on the clustering algorithm/operator that you use, if it is using value-based similarity, therefore, you need to convert your categorical data into numeric. Perhaps share your .rpm and data files for more accurate solution? I am simply guessing here.
