A Universal Similarity Model for Transactional Data Clustering
Main Article Content
Abstract
Data mining methods are used to extract hidden knowledge from large database. Data partitioning methods are used to group up the relevant data values. Similar data values are grouped under the same cluster. K - means and Partitioning Around Medoids (PAM ) clustering algorithms are used to cluster numerical data. Distance measures are used to estimate the transaction similarity. Data partitioning solutions are identified using the cluster ensembl e models . The ensemble information matrix presents only cluster data point relations. Ensembles based clustering techniques produces final data partition based on incomplete information. Link - based approach improves the conventional matrix by discovering unknown entries through cluster similarity in an ensemble. Link - based algorithm is used for the underlying similarity assessment. Pairwise similarity and binary cluster association matrices summarize the underlying ensemble information. A weighted bipartite graph is formulated from the refined matrix. The graph partitioning technique is applied on the weighted bipartite graph. The Particle Swarm Optimization (PSO) clustering algorithm is a optimization based clustering scheme. It is integrated with the clu ster ensemble model. Binary , categorical and continuous data clustering is supported in the system. The attribute connectivity analysis is optimized for all attributes. Refined cluster - association matrix (RM) is updated with all attribute relationships.
Article Details
How to Cite
, M. M. M. (2014). A Universal Similarity Model for Transactional Data Clustering. International Journal on Recent and Innovation Trends in Computing and Communication, 2(1), 73–79. https://doi.org/10.17762/ijritcc.v2i1.2916
Section
Articles