Anomaly detection in streaming data: A comparison and evaluation study

Félix Iglesias Vázquez, Alexander Hartl, Tanja Zseby, Arthur Zimek

Abstract: Highlights•Comparison and study of 8 unsupervised outlier detection methods for streaming data.•Evaluation with 180 synthetic datasets and 4 real/application datasets.•Study of challenges related to space geometries, nonstationarity and concept drift.•Effect of the memory span, the use of preknowledge, and the definition of outlierness.•Two new indices to characterize datasets and select the suitable method/algorithm.