Small-World-Like Structured MST-Based Clustering Algorithm

Abstract

Graph-theoretic clustering is one method of clustering where dataset is represented with a connected undirected graph having the distance between these points as the weights of the links between them. One approach is the construction of the Minimum Spanning Tree of said graph where the connected subgraphs formed after the removal of an inconsistent edge are the clusters. However, such methods suffer with drawbacks including partitioning without sufficient evidence and robustness to outliers. Hence, this work aims to modify the Prim's MST-based clustering algorithm to produce a spanning tree of the dataset infusing the small-world network thereby invoking its properties (i.e. small mean shortest path length and high clustering coefficient) which manifest inherent or natural clustering characteristics.