Developing a User-Friendly Software for Least-Squares Clustering

This work considers the implementation of modern intelligent cluster-analysis algorithms. The emphasis is set on theoretical description of effective algorithms as well as software modules code development. The introduction states the justification of the system development, reviews some actual clustering-related investigations and exposes basic problems that are typical for traditional methods. The theoretical part is devoted to five implemented algorithms: ik-means, A-Ward, A-Ward_pb, dePDDP, BiK-MeansR. Each of them has specifics and mathematical background. The most subtle issues are illustrated with examples. To help a user pick the right algorithm the last subsection of the theoretical part explains application areas of the algorithms. Another section aims to user point of view software description. It shows the basic graphical user interface windows and describes the user-system interaction scenario. All issues related to program implementation are detailed in section \ref{sec:structure}. This section also contains the justification of the tools and libraries choice used for development. Some aspects are revealed by code snippets. Complicated calculations are represented by flowcharts. To demonstrate how the system works we investigate a real-data example. The example involves main stages of system operation as data formation, clustering and results interpretation. Finally, the conclusion summarizes results of the work and chalks out the possible system enhancement vectors.