Abstract

In the new global economy, cyber-attacks have become a central issue. The detection, mitigation and attribution of such cyber-attacks require efficient and practical techniques to fingerprint malicious IP traffic. By fingerprinting, we refer to: (1) the detection of malicious network flows and, (2) the attribution of the detected flows to malware families that generate them. In this thesis, we firstly address the detection problem and solve it by using a classification technique. The latter uses features that exploit only high-level properties of traffic flows and therefore does not rely on deep packet inspection. As such, our technique is effective even in the presence of encrypted traffic. Secondly, whenever a malicious flow is detected, we propose another technique to attribute such a flow to the malware family that generated it. The attribution technique is built upon k-means clustering, sequence mining and Pushdown Automata (PDAs) to capture the network behaviors of malware family groups. Indeed, the generated PDAs are actually network signatures for malware family groups. Our results show that the proposed malicious detection and attribution techniques achieve high accuracy with low false (positive and negative) alerts.