Fast density estimation for density-based clustering methods

Density-based clustering algorithms are widely used for discovering clusters in pattern recognition and machine learning. They can deal with non-hyperspherical robust to outliers. However, the runtime of density-based is heavily dominated by neighborhood finding density estimation which time-consuming. Meanwhile, traditional acceleration methods using indexing techniques such as KD-tree may not be effective when dimension data increases. To address these issues, this paper proposes a fast range query algorithm, called Fast Principal Component Analysis Pruning (FPCAP), help principal component analysis technique conjunction geometric information provided attributes data. Based on FPCAP, framework accelerating developed successfully applied accelerate Density Spatial Clustering Applications Noise (DBSCAN) algorithm BLOCK-DBSCAN improved DBSCAN (called IDBSCAN) BLOCK-IDBSCAN) then obtained, respectively. IDBSCAN BLOCK-IDBSCAN preserve advantage BLOCK-DBSCAN, respectively, while greatly reducing computation redundant distances. Experiments seven benchmark datasets demonstrate that proposed improves computational efficiency significantly.