Pseudo semi-supervised general type-ii fuzzy clustering

As a growing learning philosophy, semi-supervised clustering has become widely used based on its efficacy and overwhelming superiority over unsupervised techniques. Semi-supervised clustering methods combine both supervised and supervised clustering characteristics and take advantage of their merits. This learning approach has been proved to be robust when dealing with a set of data vectors whose labels are just partly known a priori. On the other hand, based on superior uncertainty handling characteristics of General Type-2 Fuzzy sets (GT2 FSs), their application in various computational intelligence fields is growing. In real world data analysis problems, label of each data point in a dataset are mostly unknown. Based on features of semi-supervised clustering algorithms and powerful-plane representation of GT2 FSs, this paper aims to present a pseudo semi-supervised algorithm for clustering datasets with unknown labels. By saying pseudo, we mean no prior information is provided for the dataset however important data points can still be identified using a novel approach based on GT2 FSs in order to guide the clustering operation in the right direction. Extensive numerical experiments on a real gene expression dataset demonstrate effectiveness of the proposed algorithm in contrast with several other state-of-the-art fuzzy clustering techniques.