A robust unsupervised anomaly detection framework

摘要

Anomaly detection plays an essential role in monitoring dependable systems and networks such as computer clusters, water treatment systems, sensor networks, etc. However, anomaly detection nowadays remains a big challenge since previous researches suffer from inaccessible anomaly labels and inconsistent data types. Therefore, we propose a robust unsupervised anomaly detection framework (RUAD) to tackle the above problems. RUAD combines a deep AutoEncoder and a robust layer to extract the latent representations of data and separate normal data from abnormal data respectively, then utilizes Gaussian Mixture Model (GMM) to learn the distribution of normal data. In addition, our model can adapt to different types of data by simply modifying the structure of the deep AutoEncoder. Extensive experiments show that RUAD outperforms state-of-art anomaly detection techniques.