Eureka!: an interactive and visual knowledge discovery tool

Abstract

Visualization techniques may guide the data mining process since they provide effective support for data partitioning and visual inspection of results, especially when high dimensional data sets are considered. In this paper we describe Eureka!, an interactive, visual knowledge discovery tool for analyzing high dimensional 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 users to partition the search space by choosing shape or density according to criteria they deem 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 different techniques (e.g. a decision tree classifier) included in the mining tool.