Integrated Framework Using Frequent Pattern for Clustering Numeric and Nominal Data Sets

Abstract

Clustering is an exploratory technique in data mining that aligns objects which have a maximum degree of similarity in the same group. The real-world data are usually mixed in nature, i.e., it can contain both numeric and nominal data. Performance degradation is a major challenge in existing mixed data clustering due to multiple iterations and increased complexities. We propose an integrated framework using frequent pattern analysis, frequent pattern-based framework for mixed data clustering (FPMC) algorithm, to cluster mixed data in a competent way by performing a one-time clustering along with attribute reduction. This algorithm comes under divide-and-conquer paradigm, with three phases, namely crack, transformation, and merging. The results are promising when the algorithm is applied on benchmark datasets.