SDCOR: Scalable density-based clustering for local outlier detection in massive-scale datasets

This paper presents a batch-wise density-based clustering approach for local outlier detection in massive-scale datasets. Unlike the well-known traditional algorithms, which assume that all data is memory-resident, our proposed method scalable and processes input chunk-by-chunk within confines of limited memory buffer. A temporary model built at first phase; then, it gradually updated by analyzing consecutive loads points. Subsequently, end clustering, approximate structure original clusters obtained. Finally, another scan entire dataset using suitable criterion, an outlying score assigned to each object called SDCOR (Scalable Density-based Clustering Outlierness Ratio). Evaluations on real-life synthetic datasets demonstrate has low linear time complexity more effective efficient compared best-known conventional methods, need load into memory; also, some fast distance-based can perform resident disk.