Author(s)

A. Salazar, J. Gosalbez & I. Bosch

Abstract

A comparison of mining association rules from clusters generated by qualitative clustering and clusters obtained by quantitative clustering is presented. Whereas in quantitative clustering only numerical data are included, numerical and categorical data are used in qualitative clustering for record conglomeration. The aim of this paper is to compare the performance of the two different kinds of clustering by analyzing the rules obtained in a real data case from the academic area. Computational efficiency of the algorithms is compared and the reliability of the rules has been assessed by experts. For qualitative clustering three approaches have been applied: Mixed metrics for weighing quantitative and qualitative variables, symbolic clustering and conjunctive conceptual clustering. For quantitative clustering the fuzzy c-means algorithm has been applied. All the clustering methods are arranged in a hierarchical agglomerative scheme, except the conceptual clustering algorithm. Association rules have been extracted from the C4.5 decision tree algorithm. Keywords: qualitative clustering, decision rules, mixed metrics, knowledge discovery. 1 Introduction Most of the real applications include qualitative and quantitative variables for describing its data. The processing of data sets described by mixed or heterogeneous variables requires a special treatment because the standard clustering focuses on working with quantitative variables only. This article presents a study on the use of qualitative or quantitative clustering in mining association rules. Two possibilities in a data mining process are

Keywords

qualitative clustering, decision rules, mixed metrics, knowledge discovery.