An investigation of effects on cluster analysis of distance measurements and improvement

Cluster analysis is a multivariate statistical analysis that widely used in many different disciplines and with a wide range of literature. In this study, the effects of the distance measurements on cluster analysis will be investigated. In this study, our goal is clustering techniques to increase performance when the use of appropriate distance measures in clustering. To study the effects on cluster analysis of distance measures used in both simulation and real data sets. k-means algorithm is the most well known and the fast methods in non-hierarchical cluster algorithms. Because of the simplicity of k-means algorithm, this algorithm is used in various fields. Also in this study, a new method is developed for choosing the number of clusters and an algorithm to compute initial cluster centers for k-means algorithm. Moreover in this study we also showed that the effectiveness of clustering high dimensional data using principal components instead of original variables and the effectiveness of clustering high dimensional data using standardized variables instead of original variables