Optimizing the minimum spanning tree-based extracted clusters using evolution strategy

Abstract

There are many approaches available for extracting clusters. A few are based on the partitioning of the data and others rely on extracting hierarchical structures. Graphs provide a convenient representation of entities having relationships. Clusters can be extracted from a graph-based structure using minimum spanning trees (MSTs). This work focuses on optimizing the MST-based extracted clusters using Evolution Strategy (ES). A graph may have multiple MSTs causing varying cluster formations based on different MST selection. This work uses (1+1)-ES to obtain the optimal MST-based clustering. The Davies–Bouldin Index is utilized as fitness function to evaluate the quality of the clusters formed by the ES population. The proposed approach is evaluated using eleven benchmark datasets. Seven of these are based on microarray and the rest are taken from the UCI machine learning repository. Both, external and internal cluster validation indices are used to evaluate the results. The performance of the proposed approach is compared with two state-of-the-art MST-based clustering algorithms. The results support promising performance of the proposed approach in terms of time and cluster validity indices.