Title: IRIS revisited: A comparison of discriminant and enhanced rough set data analysis
Authors: Ciarán Browne, School of Information and Software Engineering , University of Ulster
Ivo Düntsch , Dept of Computer Science , Brock University , St Catherines, Ontario, L2S 3A1, Canada
Günther Gediga , Institut für Evaluation und Marktanalysen;
(Equal authorship implied)
Status: In: Rough sets in knowledge discovery, Vol.2 (Ed. Lech Polkowski and Andrzej Skowron), Physica-Verlag (1998), 345-368
Abstract: In this paper, we use the famous IRIS data set to compare the enhanced rough set analysis ROUGHIAN with Fisher's discriminant analysis method, and exhibit some general principles regarding the power of the two approaches. We show that the combination of filtering and significance testing achieves the same combination of variables in which the discriminant analysis results, with approximately the same coverage in terms of posterior probabilities. The use of significance and entropy procedures as additional information providers significantly simplifies model selection within RSDA, and justifies the appropriate choice. It turns out that prediction using the ROUGHIAN model is as good as that of discriminant analysis,

View technical report version