An Overview of Hierarchical and Non-hierarchical Algorithms of Clustering for Semi-supervised Classification

Abstract

An overview of a variety of methods of agglomerative hierarchical clustering as well as non-hierarchical clustering for semi-supervised classification is given. Two different formulations for semi-supervised classification are introduced: one is with pairwise constraints, while the other does not use constraints. Two methods of the mixture of densities and fuzzy c-means are contrasted and their theoretical properties are discussed. A number of agglomerative hierarchical algorithms are then discussed. It will be shown that the single linkage has different characteristics when compared with the complete linkage and average linkage. Moreover the centroid method and the Ward method are discussed. It will also be shown that the must-link constraints and the cannot-link constraints are handled in different ways in these methods.