Comparison of Different Hierarchical Implementations in Topological Clustering with a Hierarchical Structure

Abstract

A topological clustering algorithm can adaptively generate networks consisting of nodes and edges to approximate the data distribution. In our previous study, we proposed a topological clustering algorithm with a hierarchical structure to extract a hierarchical structure of the entire data while performing clustering. However, this algorithm does not utilize the information of clusters because each node holds the data that contributed to the creation of the node, and the data is independently used as the training data for the next layer. For improving clustering performance, this paper proposes an approach to aggregate the data that contributed to the creation of nodes in the same cluster and to utilize the aggregated data as training data for the next layer. Based on experimental results using artificial and real-world datasets, the characteristics of the proposed algorithm are discussed.