A Top-Down Approach for Hierarchical Cluster Exploration by Visualization

With the much increased capability of data collection and storage in the past decade, data miners have to deal with much larger datasets in knowledge discovery tasks. Very large observations may cause traditional clustering methods to break down and not be able to cope with such large volumes of data. To enable data miners effectively detect the hierarchical cluster structure of a very large dataset, we introduce a visualization technique HOV3 to plot the dataset into clear and meaningful subsets by using its statistical summaries. Therefore, data miners can focus on investigating a relatively smaller-sized subset and its nested clusters. In such a way, data miners can explore clusters of any subset and its offspring subsets in a top-down fashion. As a consequence, HOV3 provides data miners an effective method on the exploration of clusters in a hierarchy by visualization.