MulSim: A Novel Similar-to-Multiple-Point Clustering Algorithm

Abstract

Finding clusters in datasets with different distributions and sizes is challenging when clusters are of widely various shapes, sizes, and densities. Based on a similar-to-multiple-point clustering strategy, a novel and simple clustering algorithm named MulSim is presented to address these issues in this paper. MulSim first defines a new distance which can automatically adapt different densities when clustering. Then, the MulSim groups two points together if and only if one point is similar to another point and its similar neighbors. Our comprehensive experiments on both multi-dimensional and two dimensional datasets representing different clustering difficulties, show that the MulSim performs better than classical and state-of-the-art baselines in most cases. Besides, when increasing the size of datasets, MulSim can still ensure good clustering quality. In addition, the impact of the two MulSim parameters on clustering quality as well as the way of the parameter estimation are analyzed. In the end, the practicability and feasibility of the algorithm are tested through a face recognition example.

Keywords

• Clustering algorithm
• distance-based clustering
• similarity