Spectral Clustering Technique for Classifying Network Attacks

The problem of differentiating regular from anomalous traffic has been studied extensively. However, the classification of anomalous traffic to different types of attacks remains a difficult and widely unexplored area. In the age of big data analytics it is becoming paramount to automate the security process, such as reaction to attacks and mitigation based on their type. This paper investigates the use of spectral clustering techniques for classifying computer network attacks. Spectral clustering is a highly robust classifier for big data, and is found to accurately and efficiently classify the attack data using a minimal number of select features. Extensive investigation into feature selection and weighting is discussed. Our classification results can easily be adapted by an Intrusion Detection System (IDS) for real-time attack detection, and the classification information used to mitigate future attacks.