A self-adaptive graph-based clustering method with noise identification

Graph-based clustering methods offer competitive performance in dealing with complex and nonlinear data patterns. The outstanding characteristic of such methods is the capability to mine the internal topological structure of a dataset. However, most graph-based clustering algorithms are vulnerable to parameters. In this paper, we propose a self-adaptive graph-based clustering method (SAGC) with noise identification based on directed natural neighbor graph to auto identify the desired number of clusters and simultaneously obtain reliable clustering results without prior knowledge and parameter setting. This method adopts parameter adaptive process to deal with specific data patterns and can identify clusters with diverse shapes and detect noises. We use synthetic and UCI real-world datasets to prove the validity of the innovatory method by comparing it to k-means, DBSCAN, OPTICS, AP, SC, CutPC, and WC algorithms in terms of clustering Accuracy, Adjusted Rand index, Normalized Mutual Information and Fowlkes-Mallows index. The experimental results confirm that the proposed method contributes to the progress of graph-based clustering algorithms.