On the Detectability of Weak DoS Attacks in Wireless Sensor Networks

摘要

Anomaly detection can be used to detect malicious attacks in wireless sensor networks. Because of the autonomous nature of many sensor network scenarios, it is desirable to use an auto-configuring anomaly detector. To accomodate the indeterministic wireless environment, the anomaly detection must also be able to adapt to changing conditions. This work presents a dynamic learning extension for a pre-exisiting, autoconfiguring anomaly detection system for WSNs. Adaptability always results in possibilities for an attacker to exploit the adapting system; however, no quantitative analysis of this has been yet performed in a WSN scenario. This paper explores the trade-off between adaptability and exploitability of the system. The paper also shows that adaptability does not guarantee better false positive error rates for all attacks.