A Neural Network based Cluster Ensemble Approach for Biological Data Clustering

Abstract

Cluster Ensemble has been bloomed recently as a dominant substitute for analyzing standard cluster analysis, combining several base clustering to produce a single clustering solution, with exact accuracy, enhanced robustness and stability. Most of the earlier works implements a single clustering algorithm to perform cluster results from a large enormous amount of data. However, no single clustering method tends to be the most efficient in producing exact solutions. Several traditional Cluster Ensemble methods have been made to solve the problem of extracting exact clustering results, but it is investigated that those techniques unfortunately promotes ultimate data partition based on incomplete information. This paper exposes a new Neural Network based Cluster Ensemble approach in which it improves the similarity matrix by discovering unknown entries through proximity between the clusters in an ensemble. The Neural Network based unsupervised learning analysis is highly optimized mainly due to the estimation of distance of the weighted vector formed in Neural Network. Hence the clusters formed in the network optimally react to the certain symptoms of similarity patterns as well as the dissimilarity patterns. In particular, an efficient Neural Cluster Quality algorithm is proposed for measuring the underlying similarity estimation. Afterward, to extract the final clustering solution, a Pairwise Similarity Consensus method is used in which an effective K-Means clustering algorithm is applied over the similarity measures that are originated from the refined similarity matrix. Experimental results projected on Biological datasets retrieved from UCI repositories outperform the traditional Cluster Ensemble methods.