A visual method for high-dimensional data cluster exploration

Visualization is helpful for clustering high dimensional data. The goals of visualization in data mining are exploration, confirmation and presentation of the clustering results. However, the most of visual techniques developed for cluster analysis are primarily focused on cluster presentation rather than cluster exploration. Several techniques have been proposed to explore cluster information by visualization, but most of them depend heavily on the individual user's experience. Inevitably, this incurs subjectivity and randomness in the clustering process. In this paper, we employ the statistical features of datasets as predictions to estimate the number of clusters by a visual technique called HOV3. This approach mitigates the problem of the randomness and subjectivity of the user during the process of cluster exploration by other visual techniques. As a result, our approach provides an effective visual method for cluster exploration.