An automatic and self-adaptive multi-layer data fusion system for WiFi attack detection

Wireless networks are becoming susceptible to increasingly more sophisticated threats. Most of the current intrusion detection systems (IDSs) that employ multi-layer techniques for mitigating network attacks offer better performance than IDSs that employ single layer approach. However, few of the current multi-layer IDSs could be used off-the-shelf without prior thorough training with completely clean datasets or a fine tuning period. Dempster-Shafer theory has been used with the purpose of combining beliefs of different metric measurements across multiple layers. However, an important step to be investigated remains open; this is to find an automatic and self-adaptive process of basic probability assignment (BPA). This paper describes a novel BPA methodology able to automatically adapt its detection capabilities to the current measured characteristics, without intervention from the IDS administrator. We have developed a multi-layer-based application able to classify individual network frames as normal or malicious with perfect detection accuracy.