A Parameter-Free Clustering Algorithm Based on Density Model

Jun Mu, Hongxiao Fei, Xin Dong

Abstract: As a fundamental problem in data mining, pattern recognition and machine learning, clustering algorithm has been studied for decades, and has been improved in many aspects. However, parameter-free clustering algorithms are still quite weak, which makes their potential generalization to a lot of promising applications rather difficult. A parameter-free clustering algorithm based on density model is proposed in this paper. This algorithm explores in a dynamically constructed nearest neighbor graph to detect which points are of the same density model, and then agglomerates them into the same cluster. It requires neither previously nor interactively setting of pivotal parameters via range scaling and proportional criterion technique. Its overall computational complexity is O(n log n). And the experimental results demonstrate that the proposed algorithm can correctly recognize the arbitrary shaped clusters.