Categorical Data Clustering Using Meta Heuristic Link-Based Ensemble Method

Abstract

Conventional ensemble clustering is a consensus function that fails to produce final clusters. Such poor clusters partitioning creates poor stability with reduced clustering accuracy. This motivates to improve the final clustering quality using a hybrid ensemble-based model. In this study, an optimized link-based ensemble clustering approach is proposed to refine the incomplete datasets and to refine unknown entries in categorical dataset. The proposed work uses link-based similarity measure to find the availability of unknown datasets from link network of clusters. The ensemble clustering generates a refined cluster-association matrix in the form of weighted graphs. The final cluster partitioning acquires the final clustering partitions with a refined matrix as its input that decomposes the graph into clusters. The comparison with conventional methods is made against performance metrics to evaluate the model efficacy.