Hierarchical Brushing of High-Dimensional Data Sets Using Quality Metrics

Abstract

In this paper, we present an interactive exploration framework that puts the human-in-the-loop with the application of quality metrics and brushing techniques for an efficient visual analysis of high-dimensional data sets. Our approach makes use of the human ability to distinguish interesting structures even within very cluttered projections of the data and uses quality metrics to guide the user towards such promising projections which would otherwise be difficult or time-consuming to find. Brushing the data creates new subsets that are ranked again using quality metrics and recursively analyzed by the user. This creates a human-in-the-loop approach that makes use of hierarchical brushing and quality metrics to support interactive exploratory analysis of high-dimensional data sets. We apply our approach to synthetic and real data sets, demonstrating its usefulness.