Interactive visual mining of massive multidimensional datasets

Recent extensions to the TempleMVV multidimensional visualization system (U.S. Patent No. 5228,119) greatly enhance its analysis and data mining capabilities. The system can now be used to visually examine datasets containing hundreds of millions of records and thousands of variables in a highly interactive fashion. One can now perform fits and view the original data, fit predictions and residuals in a space of up to 10 dimensions. Two, three or thousands of dependent variables (response variables or measures) can be plotted in a 10 dimensional manner against independent variables (explanatory variables or dimensions). Boolean operator criteria can be set that control which dependent variables or sets of dependent variables are displayed in the multidimensional space. One can visually detect highly conditional domain specific correlations and associations, multidimensional filamentary clusters and isolated cells or regions of abnormally high frequency. If suitable graphic data representations are used, human pattern recognition skills allow one to quickly discover effects that often elude algorithms. Moreover the multidimensional plotting capabilities allow one to view isolated "nuggets" of information in context so that one can ascertain not only their statistical significance but also their significance relative to other "nuggets".