Combining Neuro-Fuzzy Classifiers for Improved Generalisation and Reliability.

Abstract

In this paper a combination of neuro-fuzzy classifiers for improved classification performance and reliability is considered. A general fuzzy min-max (GFMM) classifier with agglomerative learning algorithm is used as a main building block. An alternative approach to combining individual classifier decisions involving the combination at the classifier model level is proposed. The resulting classifier complexity and transparency is comparable with classifiers generated during a single crossvalidation procedure while the improved classification performance and reduced variance is comparable to the ensemble of classifiers with combined (averaged/voted) decisions. We also illustrate how combining at the model level can be used for speeding up the training of GFMM classifiers for large data sets.