Principal component analysis approach in selecting type-1 and type-2 fuzzy membership functions for high-dimensional data

With increased interest in learning from data, algorithms that manipulate datasets containing hundreds of features have become popular in various fields such as medicine, image processing, geolocation, biochemistry, and computational linguistics. Since a number of these applications exploit the power of fuzzy sets in representing uncertainties, it may be considered essential to describe a method for selecting the most suitable fuzzy membership function to represent a high-dimensional dataset. In this paper, we propose such a method, which is based on dimensionality reduction using the Principal Component Analysis (PCA) technique, followed by the Wilcoxon Minimal Bin Size algorithm, which has earlier been evaluated on multidimensional datasets up to 8 dimensions. We further demonstrate our proposed method using two real datasets consisting of 281 and 500 features, respectively.