Intelligent Anomaly Detection System for Critical Network Infrastructure

With the development of ICT technology, attempts to apply artificial intelligence to cyber security are increasing. The most relevant applications are intrusion detection systems. Traditionally, the intrusion detection system is based on known signatures. However, as attack techniques become more sophisticated, detecting unknown attacks is becoming an important issue. It is especially needed for critical network infrastructure where network reliability is crucial. Therefore, in this paper, we describe an Intelligent Anomaly Detection System(IADS) that learns the normal state to find and remove abnormalities different from the normal state. First, features are extracted from network traffic. Next, an anomaly detection function using an Autoencoder, an unsupervised learning-based algorithm learns without labels and discriminates the traffic. We validate the performance of developed systems by generating labeled datasets. Anomaly detection results are notified to the network management system so that the infrastructure can remain secure.