A New Density Clustering Method based on Dynamic Local Density

Abstract

Density-based clustering is an important research direction of data mining because density-based clustering algorithms can find clusters with arbitrary shape and are robust to noises. Local density estimation is widely used in the task of clustering and outlier detection. Traditionally, distance-based and statistic-based local density estimation methods are usually adopted for data mining. However, the effectiveness and stability of local density estimation models for clustering are still to be improved. In this paper, we propose a novel density clustering method based on dynamic local density estimation. First, we show Poisson distribution to fit the distribution of reverse <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$k$</tex> nearest neighbor counts and describe the model of dynamic local density estimation. Second, we construct a cluster order based on dynamic local density. Finally, decision graph is developed for density clustering and the identified break points partition the cluster order into clusters. Experiment results show that the new dynamic local density estimation model is effective and can help improve the performance of density clustering.