Visualization of high-dimensional model characteristics

Abstract

Using inductive learning techniques to construct explanatory models for large, high-dimensional data sets is a useful way to discover useful information. However, these models can be difficult for users to understand. We have developed a set of visualization methods that enable a user to evaluate the quality of learned models, to compare alternative models, and identify ways in which a model might be improved We describe the visualization techniques we have explored, including methods for high-dimensional data space projection, variable/class correlation, instance mapping, and model sampling We show the results of applying these techniques to several models built from a benchmark data set of census data.