Abstract:

It is well established that the programs for simulation and visualization of algorithms are excellent tools to facilitate students the understanding of them. In this paper, one of such programs is presented. It is focused on the visualization of spectral clustering algorithms. Spectral clustering is a technique that has aroused great interest because, unlike others, makes no assumption about the distribution of the clusters that are to be identified. Although the concepts used by this technique may be for the students difficult to understand, the proposed tool try to fix this problem. Being developed in Java is cross platform and the students only needs to have the Java virtual machine installed in their PCs, and in most cases is already installed. The application allows visualization of various spectral clustering algorithms and different types of connectivity graphs used in these algorithms, as well as the final found clustering. Student can load a previously saved dataset, or interactively create one, and apply the algoritms on it.

Keywords:

Algorithm visualization, data mining, spectral clustering.