Risk management in VoIP infrastructures using support vector machines

Telephony over IP is exposed to multiple security threats. Conventional protection mechanisms do not fit into the highly dynamic, open and large-scale settings of VoIP infrastructures, and may significantly impact on the performance of such a critical service. We propose in this paper a runtime risk management strategy based on anomaly detection techniques for continuously adapting the VoIP service exposure. This solution relies on support vector machines (SVM) and exploits dynamic security safeguards to reduce risks in a progressive manner. We describe how SVM parameters can be integrated into a runtime risk model, and show how this framework can be deployed into an Asterisk VoIP server. We evaluate the benefits and limits of our solution through a prototype and an extensive set of experimental results.