Analysis of CPU and GPU performances of clustering algorithms

The advancement of technology increases the competition in technology area day by day. As technology advances, it becomes more difficult to satisfy customers and users. The amount of data produced with the help of various technological devices is increasing, which causes companies to turn to different methods to analyze the data at hand. Since the analysis and interpretation of data is very important in today's world, the need and need to have this process done by machines has arisen instead of doing it manually. In cases where the labels of the available data are not known, clustering algorithms can be used to analyze them. With the help of clustering algorithms, the data can be grouped and made easier to analyze and interpret on this occasion. In this thesis, the performances of five different clustering algorithms currently used on CPU and GPU were investigated and experimental studies were carried out to detect them. In order to measure the performance of clustering algorithms, a dataset consisting of e-mails that name is Enron was used in experimental studies. As clustering algorithms in the study; model based Cobweb, density based Dbscan, grid based Clique, segmented K-Means, hierarchical Birch were selected. The necessary environment for the experimental studies was carried out on Google Colab in Python language. Experimental study results are expressed with graphs and tables, and analysis results are presented.