IMPACT OF PARAMETERS CHARACTERIZING CLUSTERING ON DATA ANALYSIS RESULTS

Abstract

Clustering algorithms are used to group some given objects defined by a set of numerical properties in such a way that the objects within a group are more similar than the objects in different groups. All clustering algorithms have common parameters the choice of which characterizes the effectiveness of clustering. The most important parameters characterizing clustering are: metrics (the distance between cluster elements and cluster centre), number of clusters k and cluster validity criteria. The goal of the paper – to perform the evaluation of the validity of metrics’ choice, to describe the change with respect to the number of clusters for experimental data purposes and to evaluate the credibility of clustering results. As an input data the table describing the rating of Latvian state higher educational institutions for year 2011 has been used and the goal of the experiment was to show, how by using the clustering methods it is possible to analyze the mentioned data in an alternative way.