M-FastMap: A Modified FastMap Algorithm for Visual Cluster Validation in Data Mining

This paper presents M-FastMap, a modified FastMap algorithm for visual cluster validation in data mining. In the visual cluster validation with FastMap, clusters are first generated with a clustering algorithm from a database. Then, the FastMap algorithm is used to project the clusters onto a 2-dimensional (2D) or 3-dimensional (3D) space and the clusters are visualized with different colors and/or symbols on a 2D (or 3D) display. From the display a human can visually examine the separation of clusters. This method follows the principle that if a cluster is separate from others in the projected 2D (or 3D) space, it is also separate from others in the original high dimensional space (the opposite is not true). The modified FastMap algorithm improves the quality of visual cluster validation by optimizing the separation of clusters on the 2D or (3D) space in the selection of pivot objects (or projection axis). The comparison study has shown that the modified FastMap algorithm can produce better visualization results than the original FastMap algorithm.