Search Space Partitioning to Enhance Outlier Rule Discovery
Michal J. Okoniewski, Piotr Gawrysiak, Łukasz Gancarz · 2002
This paper presents an enhancement of the multidimensional quantitative rule discovery methodology presented in [9] by clustering of preselected μ-tuples. This unconventional, in KDD cycle, usage of clustering as a preprocessing step allows for significant speed-up without compromising informational value of discovered results. In that sense it is a continuation of research described in [2]. The paper presents the outcome of experiments performed with this method over datasets obtained by random sampling of graphic image files. The outcome clearly shows the advantages and applicability of proposed methodology.