Higher-Order PCA for anomaly detection in large-scale networks

Anomaly detection is important to monitor and keep the health of large scale IP networks. principal component analysis (PCA) based methods have been proposed with major limitation on the scalability. In this paper, we apply higher-order singular value decomposition (HOSVD) and higher-order orthogonal iteration (HOOI) algorithms on network traffic anomaly detection by rearranging the data in tensor formats. Also a low-complexity implementation of the HOOI algorithm is developed. Simulation results show that the higher-order methods improve the detection performance and also reduce the complexity for large-scale networks.