Network anomaly detection based on tensor decomposition

The problem of detecting anomalies in time series from network measurements has been widely studied and is a topic fundamental importance. Many anomaly detection methods are based on the inspection packets collected at core routers, with consequent disadvantages terms computational cost privacy. We propose an alternative method which packet header not needed. extraction normal subspace obtained by tensor decomposition technique considering correlation among metrics. In its online version, proposed approach for allows efficient tracking changes subspace. flexibility illustrated applying it to distinct examples that include supervised unsupervised detection. use actual data residential routers.