A Comprehensive Study of Ensemble Models to Improve the Performance of Cluster Algorithms

The study analyzed the individual performance of partition cluster algorithms and selected Kmeans, Kmeans 9+ , Kmedoid, and Fuzzy Cmeans algorithms as base algorithms for the ensemble.The cluster performance is assessed using UCI data sets as well as other common public data sets.The quality of cluster results depends on the base cluster algorithm used.The efficiency of base algorithms is added based on the ensemble models.We developed two ensemble models: a simple hard voting ensemble and a soft boosting ensemble based on the bagging and boosting ensemble technique.Ensemble of different cluster algorithms can generate the most accurate clusters.Both models show better cluster results than their base cluster algorithms for the small and big data sets.When using most data sets, the Soft Boosting Ensemble model achieves 100% cluster accuracy.The cluster evaluating functions are the benchmark for assessing the quality of the cluster.All the cluster evaluating indices show better performance for developed ensemble models.Internal cluster-evaluating indices as well as external cluster-evaluating indices are used to compare the cluster quality of the individual cluster algorithm and generated ensemble cluster models.The work establishes that the developed ensemble methods improved the quality of the generated clusters.