Improve neuro-fuzzy learning by attribute reduction

Neuro-fuzzy learning is a combination of neural networks and fuzzy systems to learn fuzzy rules from examples. One of the popular tools for neuro-fuzzy learning is the adaptive network based fuzzy inference systems (ANFIS) introduced by Jang. It is observed from our past experiments that data sets with more than six attributes (features) may present a challenge to ANFIS learning. Rough set theory introduced by Pawlak has been shown as an effective tool for data reduction. This paper studied how ANFIS learning may benefit from using rough set tools for data reduction. Empirical results show that ANFIS learning from reduced data sets usually has better prediction accuracies and faster learning time.