Comparison of partitioning-based clustering algorithms on differently distributed data

As a result of wide spread technology usage, large volumes of collected data began to emerge. It is impossible to discover and analyze any information in such large data collection, so data mining comes into play. Data mining is a process that discovers unpredictable and usable knowledge from databases. In other words, data mining is the process of finding relation patterns, changes, deviations and trends, as well as interesting information like specific structures from large databases. One of the widely used data mining methods is clustering, which divides the data set into different clusters while trying to make the likelihood ratio as minimum inside the cluster and as maximum among other clusters depending on the options in the database. In this study, partitioning-based clustering methods are compared by applying them on data sets with different distribution patterns. We used k-means and kernel k-means partitioning algorithms for clustering data sets. By applying clustering operations on differently distributed data sets we compared the speed, clustering quality and the size of memory used in clustering for these algorithms. The information that we gathered by this comparison is presented and discussed in the related sections of this thesis.