COMPUTATIONAL SEMANTIC REPRESENTATION GUARANTEES INTERPRETABILITY OF FUZZY RULE BASED CLASSIFIER

Abstract

The fuzzy rule-based classifier design methods have been widely studied by the research community due to many practical applications in the real life. The quality of a classifier clearly depends on the semantic representations of linguistic words in the rule bases. Hedge algebra allows to the creation of a formal formalism for designing the fuzzy sets-based computational semantics of linguistic words from their inherent semantics. However, the existing design methods of fuzzy sets-based computational semantics of linguistic words do not guarantee the interpretability of the fuzzy rule-based classifiers. Specifically, the designed multi-granularity representation does not retain the generality-specificity relation of linguistic terms. This paper presents a fuzzy sets-based computational semantic representation that guarantees the interpretability of the fuzzy rule-based classifier. Experimental results on 23 real-world datasets have shown that the proposed method gives better classification accuracy while not increasing the complexity of the fuzzy rule-based systems in comparison with the existing methods.