Evolving Data Monitoring Algorithm based on Data Stream Clustering

Abstract

Clustering is one of the most important methods in the data stream mining. The nature of evolving data streams puts forward the following requirements for clustering: discovery of clusters with arbitrary shape, determination of the optimal number of clusters, anomaly detection and processing the influence of time. In the communication process, timely detection of anomalies is very important for the coordination of various devices. In this paper, we propose an adaptive density-based approach for clustering in evolving data streams(SagsDStream), which automatically updates parameters in subsequent stages based on the offline clustering result, so that the parameters can adapt to the change of data stream. In the online mode, the judging rules are saved in Improved Time Cluster Feature (ITCF). In the offline model, we use the density of the data distribution and the cluster distance to set the parameters of clustering algorithm, which overcome the sensitivity of Density-Based Spatial Clustering of Applications with Noise (DBSCAN) threshold setting. Experimental results show that SagsDStream has better processing efficiency and clustering effect when dealing with data streams.