Skip to Main Content
The strength of neuro-fuzzy systems involves two contradictory requirements in neuro-fuzzy modeling: interpretability versus accuracy. The Yager-inference-scheme-based fuzzy CMAC (FCMAC-Yager) architecture shows advantages such as it exhibits learning and memory capabilities of the human cerebellum through the CMAC (cerebellar model articulation controller) structure and the human way of reasoning through the Yager inference scheme. However, it suffered from an exponential increase in the number of identified fuzzy rules and computational cost arising from high-dimensional data. This diminishes the interpretability of the FCMAC-Yager network in linguistic fuzzy modeling. This paper proposes a novel rough set-based rule reduction (RSFCMAC) approach for the established FCMAC-Yager architecture. RSFCMAC algorithm used in the FCMAC-Yager network can help to provide better generalization, to reduce the number of fuzzy rules and computational cost. The proposed algorithm not only performs reduction of redundant fuzzy rules, but also carries out reduction of redundant input attributes. Experiments using real-world application involving stock movement and highway traffic flow prediction were conducted to evaluate the performance of the proposed RSFCMAC against the FCMAC-Yager network and other published results of cross-architectures using globalized learning as well as similar architectures employing localized learning. The results are encouraging.