A Visual Approach for Classification Based on Data Projection

In this paper we present a visual approach for classification in data mining, based on the enhanced separation feature of a visual technique, called Hypothesis-Oriented Verification and Validation by Visualization (HOV3). In this approach, the user first projects a labeled dataset by HOV3with a statistical measurement of the dataset on a 2d space, where data points with the same class label are well separated into groups. Then each well separated group and its measure vector are employed as a visual classifier to classify unlabeled data points by projecting and grouping them together with the overlapping labeled data points. The experiments demonstrate that our approach is effective to assist the user on classification of data by visualization.