This paper introduces the Unsupervised Neural Netbased Intrusion Detector (UNNID) system, whichdetects network-based intrusions and attacks usingunsupervised neural networks. The system has facilitiesfor training, testing, and tunning of unsupervised nets tobe used in intrusion detection. Using the system, wetested two types of unsupervised Adaptive ResonanceTheory (ART) nets (ART-1 and ART-2). Based on theresults, such nets can efficiently classify network trafficinto normal and intrusive. The system uses a hybrid ofmisuse and anomaly detection approaches, so is capableof detecting known attack types as well as new attacktypes as anomalies.
展开▼