Linked Spectral Graph based Cluster Ensemble Approach using Weighted Spectral Quality Algorithm for Medical Data Clustering

摘要

Over the certain span of time, Cluster Ensembles have been emerged as an offspring for solving the problem of extracting the efficient clustering results. Although many efforts have been commenced, it is examined that these techniques adversely creates the final data partition based on imperfect information. The original Ensemble information matrix exposes only the cluster data object relations with many entries being left empty. This paper presents an investigation that provides a solution to the problem of degrading the quality of the final partition through a Linked Spectral Graph based Cluster Ensemble approach. In particular, an effective Weighted Spectral Quality algorithm is proposed for the underlying similarity measurement among the Ensemble Members which in turn can be highly used to avoid the local optimum and the ill-posed issues derived from the huge dimensional samples. Subsequently, to obtain the final ultimate clustering results a Spectral Clustering based Consensus Function is applied to the Distilled Similarity Matrix (DSM) that is formulated from the similarity assessment algorithm. The Experimental results projected on Medical datasets retrieved from the UCI repository demonstrate that the proposed approach outperforms the traditional ones in data clustering.