A Linguistic Fuzzy-XCS classifier system

Abstract

Data-driven construction of fuzzy systems has followed two different approaches. One approach is termed precise (or approximative) fuzzy modelling, that aims at numerical approximation of functions by rules, but that pays little attention to the interpretability of the resulting rule base. On the other side is linguistic (or descriptive) fuzzy modelling, that aims at automatic rule extraction but that uses fixed human provided and linguistically labelled fuzzy sets. This work follows the linguistic fuzzy modelling approach. It uses an extended Classifier System (XCS) as mechanism to extract linguistic fuzzy rules. XCS is one of the most successful accuracy-based learning classifier systems. It provides several mechanisms for rule generalization and also allows for online training if necessary. It can be used in sequential and non-sequential tasks. Although originally applied in discrete domains it has been extended to continuous and fuzzy environments. The proposed Linguistic Fuzzy XCS has been applied to several well-known classification problems and the results compared with both, precise and linguistic fuzzy models.