D. Marghescu (Finland)
Data treatment and visualization, evaluation, visual data mining, cluster validity
When dealing with large amounts of high-dimensional data, one approach is to reduce the number of data dimensions by applying a projection technique. The later reduces the data dimensionality by combining the original variables into a smaller number of new dimensions, in a linear or nonlinear manner. The projection methods are particularly useful because they lend themselves to visual representations of data, when the number of new dimensions is one, two or three. In this paper, we compare different visualization techniques based on projection techniques with respect to their effectiveness for solving a data-mining task such as clustering. We investigate the use of cluster validity measures in order to judge the effectiveness of the projection techniques in visual data mining. The results show that cluster validity is a successful approach to evaluate objectively the visualization techniques.
Important Links:
Go Back
