Cluster Analysis: Cluster Analysis Theory And Methods

Chapter 3: Methodology

3.1 Introduction

This chapter is presented detail description about the cluster analysis theories and methods which were used in this study. This will help to identify how those methods and theories were used in achieving objectives in this study.

3.2 Cluster Analysis

Cluster analysis can be viewed as dividing similar objects or data into categories or groups (clusters) that are meaningful, useful or both. Cluster analysis is very useful concept for data summarization. When it comes to design meaningful clusters, natural structure of data are considered. Human beings have skills for dividing objects into similar groups and assigning particular objects into those groups. Cluster analysis is applied in practical scenarios

It is a distance between a point P and distribution D and it measures number of standard deviations from point P to mean D.

Mahalanobis distance= 〖√(x-μ)〗^T S^(-1) (x-μ)

3.6 Hierarchical Methods

Hierarchical clustering can be generalized as a series of successive merges or series of successive divisions. Agglomerative method and divisive method are two form of hierarchical clustering. Agglomerative method:

In agglomerative method, initially there are many clusters because clustering starts with individual objects. That means initially objects are considered as clusters. Then the most similar objects are grouped into one cluster. Based on the similarity those groups are merged into one group. When similarities decrease those groups are finally merged into one group.

Divisive method:

Divisive method is opposite method of agglomerative method. That means initially there is one group of objects. Then it divided into two sub groups based on dissimilarity. Objects in the one sub group are far away from other sub group. In this method, those initial sub groups are divided further until one object