| Although a lot of work has been done in proposing secure mechanisms for sensor networks, very little work has been reported on how to respond to attacks that have infiltrated the network. In this thesis we suggest a response framework to attacks that have been detected. We look at one particular response in detail, namely deception. Our contributions in this thesis are twofold. (1) A time-series based deception framework is proposed for learning attacker's behaviors and deceiving the attacker. A mathematical generalization of the framework is also proposed. We simulate a denial of service attack and simulation results show that, to the attacker, genuine response data and the fake response data generated by our algorithm are highly indistinguishable. (2) A harmonic means mechanism is used to determine if the attacker has changed his behavior. Simulations are provided to show that our approach not only detects slight changes over time, but also does filter out abnormality in data samples to avoid false alarms. |
