Tuning fuzzy partitions or assigning weights to fuzzy rules: which is better?

The accuracy of linguistic classifiers can be improved with several techniques, but they all compromise the interpretability of the rule base up to a certain degree. Assigning weights to fuzzy rules and tuning the memberships associated to linguistic variables are two of the most common methods. In this work we study whether tuning the membership functions in a linguistic classifier is better or not than adjusting rule weights, in terms of the interpretability of the rule base and the complexity of the output.