TY - GEN
T1 - SPATIO-TEMPORAL ANOMALY DETECTION FOR LARGE-SCALE DYNAMIC ATTRIBUTED NETWORKS
AU - Wu, Hui
AU - Li, Yan Fu
N1 - Publisher Copyright:
© ESREL 2021. Published by Research Publishing, Singapore.
PY - 2021
Y1 - 2021
N2 - Dynamic attributed networks (DANs) provide powerful means of representing complex system, e.g., online social networks, financial networks, transactional networks, and wireless sensor networks. To facilitate situation awareness and critical decision-making, anomaly detection in DANs has become an increasingly active area of research in network sciences. However, most existing methods are only capable of detecting the temporal outliers, neglecting the potential benefits of jointly detecting the spatial outliers across the entire network. To address this issue, this paper presents a novel approach, which is also efficient for large-scale networks. Specifically, we first develop a novel recurrent neural network structure to explore the spatio-temporal correlations of the DANs. Furthermore, prediction residuals are monitored through an exponentially weighted moving average (EWMA) control chart. Experiments on synthetic and real-world datasets depict the properties and benefits of the method compared with existing methods in the literature.
AB - Dynamic attributed networks (DANs) provide powerful means of representing complex system, e.g., online social networks, financial networks, transactional networks, and wireless sensor networks. To facilitate situation awareness and critical decision-making, anomaly detection in DANs has become an increasingly active area of research in network sciences. However, most existing methods are only capable of detecting the temporal outliers, neglecting the potential benefits of jointly detecting the spatial outliers across the entire network. To address this issue, this paper presents a novel approach, which is also efficient for large-scale networks. Specifically, we first develop a novel recurrent neural network structure to explore the spatio-temporal correlations of the DANs. Furthermore, prediction residuals are monitored through an exponentially weighted moving average (EWMA) control chart. Experiments on synthetic and real-world datasets depict the properties and benefits of the method compared with existing methods in the literature.
KW - Anomaly detection
KW - Dynamic attributed networks
KW - Exponentially weighted moving average control chart
KW - Large-scale networks
KW - Recurrent neural network structure
KW - Spatio-temporal correlations
UR - https://www.scopus.com/pages/publications/85135490472
U2 - 10.3850/978-981-18-2016-8_662-cd
DO - 10.3850/978-981-18-2016-8_662-cd
M3 - 会议稿件
AN - SCOPUS:85135490472
SN - 9789811820168
T3 - Proceedings of the 31st European Safety and Reliability Conference, ESREL 2021
SP - 2932
BT - Proceedings of the 31st European Safety and Reliability Conference, ESREL 2021
A2 - Castanier, Bruno
A2 - Cepin, Marko
A2 - Bigaud, David
A2 - Berenguer, Christophe
PB - Research Publishing, Singapore
T2 - 31st European Safety and Reliability Conference, ESREL 2021
Y2 - 19 September 2021 through 23 September 2021
ER -