Non-conforming behavior detection for VoIP-based network systems

This work proposes a detection scheme that identifies non-conforming behavior in a VoIP network, based on statistical analysis and hypothesis testing. VoIP networks are a popular, low-cost alternative for telephony that offer lower rates especially for long-distance calls. Other services such as FollowMe, enhance the traditional voice-oriented nature of these networks. Consequently several security concerns such as fraud calls, are related to the high availability required by a VoIP system. Fraud calls account for an average loss of 3% to 5% of the operators' revenue. Thus the detection and prevention of the users from behaving in a non-conforming way, becomes crucial. A trustworthy and secure management and billing scheme is necessary, to guarantee the proper operation. This work proposes a behavioral control scheme for the VoIP clients. An initial training period defines the normal behavior. Then statistical analysis and t-testing is employed to extract results regarding the users' profiles, with pre-defined confidence levels. A Buffer zone creates a more flexible decision-making process. The scheme also offers the ability to configure its parameters, in order to react appropriately under different network conditions and detect possible misuses. It is implemented and its rational operation is verified via several simulation scenarios.