A Frequent and Rare Itemset Mining Approach to Transaction Clustering

Abstract

Data clustering is the unsupervised learning procedure of grouping related objects based on similarity measures. Intra cluster similarity is maximized and inter cluster similarity is minimized in the clustering technique. Distance based similarity measures are employed in k-means, k-mediods, etc. which are some of the clustering algorithms. This project explores the scope of guided clustering, wherein frequent and rare patterns shall be employed in the process of clustering. The work focuses on having a better centre of a cluster, employing variants of frequent and rare itemsets such as Maximal Frequent Itemset (MFI) and Minimal Rare Itemset (MRI). The literature supports several instance of association rule based classification and the effort is to have a MFI/MRI based clustering. The proposed model employs MFI and MRI in the process of choosing cluster centers. The proposed algorithm has been tested over benchmark datasets and compared with centroid based hierarchical clustering and large items based transaction clustering algorithms and the results indicate improvement in terms of cluster quality.