General type-2 fuzzy membership function design and its application to neural networks
Eun-A Shim, Frank Chung-Hoon Rhee · 2011
Several type-1 fuzzy membership function (Tl FMF) generation methods have been proposed to model the uncertainty of pattern data. However, if we cannot obtain satisfactory results using type-1 fuzzy sets, employment of type-2 fuzzy sets (T2 FSs) for managing uncertainty may allow us to obtain desirable results. In this paper, a general T2 FMF design method and its application to back propagation (BP) neural networks is proposed. The general T2 FMF is designed using data histograms and then type-1 fuzzy membership values which are extracted from the centroid of the T2 FMF are used as inputs to the BP neural network. Applying our proposed membership assignment to the BP neural networks shows improvement of the classification performance since the uncertainty of pattern data are desirably controlled by the T2 fuzzy memberships. Experimental results for several data sets are given.