Different Schemes for Improving Fuzzy Clustering Through Supervised Learning

Abstract

Clustering is an important unsupervised learning tool to find natural grouping of instances from a given unlabeled data based on some similarity/dissimilarity measures. However, clustering the data points at the boundary of multiple overlapping clusters is really a challenge. The challenge also lies in many real life applications where the boundaries among clusters are not rigid and clear. There are several hard and soft clustering algorithms which attempt to address these challenges. Hard clustering algorithms though perform well in the presence of well separated clusters, it shows poor performance in presence of such difficulties mentioned above. Researchers have recommended several soft clustering algorithms for handling such dilemma. In this article, we have proposed different schemes toward improvement of Fuzzy clustering employing supervised classification. The superiority of our prescribed schemes are demonstrated by comparing its performance with that of Fuzzy c-means clustering. The comparison has also been made among different schemes we have proposed in this article.