An Efficient Density-Based Clustering Algorithm for Vertically Partitioned Distributed Datasets

Clustering is an important research in data mining. Clustering massive datasets has especially been a challenge for its large scale and too much noise data points. Distributed clustering is an effective method to solve these problems. Most of existing distributed clustering research aims at circumstances of horizontally partitioned dataset. In this paper, considering vertically partitioned distributed datasets, based on the analysis of relations between local noise datasets and the corresponding global one, an efficient filtering is applied to the global noise, which can efficiently eliminate the negative affection of noise data and reduce the scale of dataset to be dealt on the center node. Furthermore, an effect storage structure CTL(closed triangle list) is designed to store the intermediate clustering results of each node, which can efficiently reduce communication costs among distributed computer nodes during the clustering process and is helpful to conveniently generate global clustering model with high space utilization ratio and complete clustering information. Thus,a distributed density-based clustering algorithm DDBSCAN is proposed. Theoretical analysis and experimental results testify that DDBSCAN can effectively solve the problem of clustering massive vertically partitioned datasets, and the algorithm is effective and efficient.