Sparse Clustering for Probability Un-Weighted Graphs Mining

-Probabilistic graphs have significant importance in data mining. The correlations endure amid the adjacent edges in different probabilistic graphs. Graph clustering is used in exploratory data analysis at data compression, information retrieval and image segmentation. The existing work presented a Partially Expected Edit Distance Reduction (PEEDR) and Correlated Probabilistic Graphs Spectral (CPGS) clustering algorithms. Pruning techniques are developed to improve the efficiency of clustering algorithms. In PEEDR algorithm, the cluster graph is constructed iteratively improve it by adding or removing vertices from some clusters. Drawback of the existing work was sparseness in cluster, that makes the correlated probability graph noisier and become difficult to solve. The tradeoff arises in minimizing errors between missing edges within clusters and present edges across clusters. A Sparse Cluster Correlation Probability un-weighted graphs model is proposed. This method improves an efficient clustering algorithm for probabilistic graphs. A combinatorial optimization problem convex relaxation of sparseness of the correlated cluster graph and cluster matrix is low-rank for random clustered graphs. This problem gets better true clustering for bigger range of sparsity and cluster sizes. Both level of sparsity and number and sizes of the clusters are allowed to be functions of the total number of nodes. The performance metrics such as number of un-weighted graph nodes, edge size, optimal error density, cluster size, and cluster generation time are evaluated. Key Terms: Sparse clustering, correlation edges, convex relaxation.