Eureka!: A tool for interactive knowledge discovery

Abstract

In this paper we describe an interactive, visual knowledge discovery tool for analyzing numerical data sets. The tool combines a visual clustering method, to hypothesize meaningful structures in the data, and a classification machine learning algorithm, to validate the hypothesized structures. A two-dimensional representation of the available data allows a user to partition the search space by choosing shape or density according to criteria he deems optimal. A partition can be composed by regions populated according to some arbitrary form, not necessarily spherical. The accuracy of clustering results can be validated by using a decision tree classifier, included in the mining tool.