Research on Data Stream Dynamic Clustering Algorithm Based on Density Peaks
The uncertainty in data stream and the challenge of identifying data of arbitrary shapes and noise in such environ-ments which have attracted widespread attention are addressed by proposing a robust density peak-based dynamic clustering algorithm for data stream. The clustering algorithm framework consists of online and offline stages,where the online stage aims to respond instantly and process continuously arriving data. It reduces the influence of historical data on clustering by designing an uneven decay strategy for micro-clusters and dynamically weighting samples based on their distances to micro-clusters. In the offline stage which is based on density peak clustering a locally adaptive density calculation method is introduced to mitigate the'domino effect 'during the density peak assignment phase. This method can handle complex data processing which is not affected by limited memory and exhibits good robustness. Comparative experiments are conducted on both artificial and real datasets,with the experimental results demonstrating the superiority of the proposed algorithm over others. The robust density peak-based dynamic clustering algorithm for data stream presented in this paper yields better clustering performance.
data streamclusteringnon-uniform decay strategynearest neighborhooddensity peaks
NETL
NSTL
万方数据
