Unsupervised Optimal Discriminant Plane Based Feature Extraction Method

Optimal discriminant plane based on Fisher criterion function is an important supervised feature extraction method and has great influence in the area of pattern recognition. In this paper, an extension of optimal discriminant plane in unsupervised pattern is presented. The basic idea is to optimize the defined fuzzy Fisher criterion function to figure out an optimal discriminant vector and fuzzy scatter matrixes. With these, a novel feature extraction method based on unsupervised optimal discriminant plane can be obtained. The experimental results for three UCI datasets in clustering validity experiments demonstrate that although this method in unsupervised pattern can not have the same performance as optimal discriminant plane feature extraction method in supervised pattern, it is superior over principal components analysis unsupervised feature extraction algorithm.