Adaptive Sampling and Learning for Unsupervised Outlier Detection.

Unsupervised outlier detection algorithms often suffer from high false positive detection rates. Ensemble approaches can be used to address this problem. This paper proposes a novel ensemble method which adopts the use of an adaptive sampling approach, and combines the outputs of individual anomaly detection algorithms by a weighted majority voting rule in a complete unsupervised context. Simulations on well-known benchmark problems show substantial improvement in performance.