Improving Projection-based Data Analysis by Feature Space Transformations

摘要

Generating effective visual embedding of high-dimensional data is difficult - the analyst expects to see the structure of the data in the visualization, as well as patterns and relations. Given the high dimensionality, noise and imperfect embedding techniques, it is hard to come up with a satisfactory embedding that preserves the data structure well, whilst highlighting patterns and avoiding visual clutters at the same time. In this paper, we introduce a generic framework for improving the quality of an existing embedding in terms of both structural preservation and class separation by feature space transformations. A compound quality measure based on structural preservation and visual clutter avoidance is proposed to access the quality of embeddings. We evaluate the effectiveness of our approach by applying it to several widely used embedding techniques using a set of benchmark data sets and the result looks promising.