Evaluative Studies of Fuzzy Knowledge Discovery through NF Systems
Arthur Ramer, M. C. Nicoletti, Sam Yuan Sung · Studies in fuzziness and soft computing · 2000
Recent years witnessed a rapid growth in fuzzy modelling using neurofuzzy systems (NF Systems or NFS for brief). Typically neural net structures would be extended to permit fuzzy inputs and to allow the outputs to be interpreted as fuzzy. The last step would defuzzify these outcomes producing a classifier, which would consist of a family of threshold (“partitioning ”) values. This chapter focuses on fuzzy modelling as a formalism for representing and inferring knowledge. It is mainly concerned with the investigation of the use of a neurofuzzy system for inducing fuzzy knowledge in the form of fuzzy if-then rules from a set of preclassified instances. We describe the experimental studies which we conducted in a few different knowledge domains. The motivation of our work is to evaluate the possibilities offered by this approach, to learn about its contributions and to establish situations where its use is appropriate. We close by discussing lines of further rearch into performance of such systems.