Densityplot matrix display for large distributed data
Jing Zhang · 2004
Data visualization techniques are essential in large data mining process because of their involvement of direct human interactions. Dynamic interaction is recognized as a critical component for very large, multidimensional data analysis. Most of the visualization tools available today can provide only static displays of large data set. We present a visualization tool, Limn Matrix, in this thesis. Limn Matrix uses scatterplot matrix display because of its power effectiveness in exploring relationships between multiple variables. Limn Matrix uses a density transformation to solve the overplotting problem associated with using scatterplots with large data sets. Overplotting counts are transformed to shades of grays in the density transformation. We developed an indexing structure for Limn Matrix to allow dynamic interactions between scatterplots. Records in a large data set are mapped to the density counts of the plots through these new indexes. By combining sampling techniques with our indices, Limn Matrix can reduce the number of data cases and provide support for real-time interaction with large, multidimensional data.