Μελέτη αλγορίθμων ομαδοποίησης σε περιβάλλον προγραμματισμού Python
Python based study of clustering algorithms
View/ Open
Abstract
Cluster analysis is the field of unsupervised learning that includes processes which divide data into groups according to a proximity measure. We briefly review the theoretical foundations of the field and provide a description of the programming concepts and tools used throughout this study. We also describe, build and use statistical techniques and indices suitable for the evaluation of clustering results. We implement seven different data clustering algorithms which can be organized into three different categories and test each one of them on three different datasets of synthetic data. In the final chapter, which can be considered a second distinctive part, we apply some of these algorithms combined together to accomplish image segmentation analysis tasks. We execute our algorithms on a set of images and measure the performance of those clustering-based segmentation results with reference to human made segmentation. We finally propose and construct a merging technique based on depth first search algorithm that when applied to an already clustered image, raises the performance dramatically.