Design and Validation of an Explainable Fuzzy Beer Style Classifier

Abstract

We describe step by step how to design, implement and validate an interpretable fuzzy rule-based beer style classifier endowed with explanation capability. First, we revise some preliminary work regarding both interpretable fuzzy modeling methodologies and related software. Second, we introduce the use case on beer style classification. Third, we build and validate a fuzzy rule-based classifier with a good interpretability-accuracy trade-off for this use case. Fourth, we endow this classifier with explanation capability through a general linguistic interface that is tuned ad-hoc for the use case under consideration. Fifth, we show how the designed explainable classifier can be refined and exploited with several interoperable software tools. Finally, we compare two kinds of multi-modal (i.e., textual and graphical) explanations: (1) explanations which are inherently natural and fully-meaningful to users, because they are supported by an interpretable fuzzy rule-based classifier which is carefully designed (i.e., it is grounded in common-sense expert knowledge and global semantics); and (2) explanations which are supported by the linguistic approximation of a fuzzy rule-based classifier extracted from data with the focus only on accuracy (thus lacking of linguistic interpretability). The use case is implemented with open source software and all related datasets, tools and scripts are available online for the sake of reproducibility.