Clustering Techniques

Abstract

Clustering is one of the most widely researched areas in unsupervised learning, where the main aim is to find structures in unlabelled data sets. This is done by partitioning data set into smaller groups or clusters so that the data points in the cluster have more common features among themselves compared to those in other clusters. There are plenty of different types of clustering techniques starting from the classical to the more recent ones based on the topological and geometrical methods. It has wide application across various fields. Different types of hierarchical, partitioning and density-based clustering algorithms are studied along with topological data analysis based clustering using persistent homology. The real data sets contain both numerical and categorical variables, which makes it difficult to cluster. Different approaches and few techniques for clustering mixed data sets are discussed. The objective is to study all these techniques and their limitations complemented by two real-life application in business and physical science fields.