Traditional Network Intrusion Detection Systems (NIDSs) rely on either specialized signatures of previously seen attacks, or on expensive and difficult to produce labeled traffic datasets for user-profiling to hunt out network attacks. Despite being opposite in nature, both approaches share a common downside: they require the knowledge provided by an external agent, either in terms of signatures or as normal-operation profiles. In this paper we present UNIDS, an Unsupervised Network Intrusion Detection System capable of detecting unknown network attacks without using any kind of signatures, labeled traffic, or training. UNIDS uses a novel unsupervised outliers detection approach based on Sub-Space Clustering and Multiple Evidence Accumulation techniques to pin-point different kinds of network intrusions and attacks such as DoS/DDoS, probing attacks, propagation of worms, buffer overflows, illegal access to network resources, etc. We evaluate UNIDS in three different traffic datasets, including the well-known KDD99 dataset as well as real traffic traces from two operational networks. We particularly show the ability of UNIDS to detect unknown attacks, comparing its performance against traditional misuse-detection-based NIDSs. In addition, we also evidence the supremacy of our outliers detection approach with respect to different previously used unsupervised detection techniques.

Abstract. Current network monitoring systems rely strongly on signature-based and supervised-learning-based detection methods to hunt out network attacks and anomalies. Despite being opposite in nature, both approaches share a common downside: they require the knowledge provided by an expert system, either in terms of anomaly signatures, or as normal-operation profiles. In a diametrically opposite perspective we introduce UNADA, an Unsupervised Network Anomaly Detection Algorithm for knowledge-independent detection of anomalous traffic. UNADA uses a novel clustering technique based on Sub-Space-Density clustering to identify clusters and outliers in multiple low-dimensional spaces. The evidence of traffic structure provided by these multiple clusterings is then combined to produce an abnormality ranking of traffic flows, using a correlation-distance-based approach. We evaluate the ability of UNADA to discover network attacks in real traffic without relying on signatures, learning, or labeled traffic. Additionally, we compare its performance against previous unsupervised detection methods using traffic from two different networks.