Performance analysis of branch-and-bound approach with various model-selection clustering techniques for image data point

Abstract

In data mining context, for efficient data analysis recent researchers utilized branch-and-bound methods such as clustering, seriation and feature selection. Traditional cluster search was done with different partitioning schemes to optimize the cluster formation. Considering image data, partitioning approaches seems to be computationally complex due to large data size, and uncertainty of number of clusters. Recent work presented a new version of branch and bound model called model selection problem, handles the clustering issues more efficiently. For model-based clustering problems, to assign data point to appropriate cluster, cluster parameters should be known. Cluster parameters are computed only if the cluster assignments are known. Data point is assigned to the cluster based on most matching model such as Navigation and Cost Model, Segment Representation in SwiftRule and Analytic model. If the problem-specific bounds and/or added heuristics in the data points of the domain area get surmounted, memory overheads, specific model selection, and uncertain data points cause various clustering abnormalities. In addition cluster validity and purity needs to be testified for efficiency of problem-specific bound on certain domain areas of image data clustering. Experimental evaluation on the model selection approach of cluster model shows the improvement in accuracy, computational complexity and execution time, when compared to Navigation and Cost Model, Segment Representation in SwiftRule and Analytic model.