Empirical Techniques To Detect Rogue Wireless Devices

Abstract

Media Access Control (MAC) addresses in wireless networks can be trivially spoofed using off-the-shelf devices. We proposed a solution to detect MAC address spoofing in wireless networks using a hard-to-spoof measurement that is correlated to the location of the wireless device, namely the Received Signal Strength (RSS). We developed a passive solution that does not require modification for standards or protocols. The solution was tested in a live test-bed (i.e., a Wireless Local Area Network with the aid of two air monitors acting as sensors) and achieved 99.77%, 93.16%, and 88.38% accuracy when the attacker is 8–13 m, 4–8 m, and less than 4 m away from the victim device, respectively. We implemented three previous methods on the same test-bed and found that our solution outperforms existing solutions. Our solution is based on an ensemble method known as Random Forests. We also proposed an anomaly detection solution to deal with situations where it is impossible to cover the whole intended area. The solution is totally passive and unsupervised (using unlabeled data points) to build the profile of the legitimate device. It only requires the training of one location which is the location of the legitimate device (unlike the misuse detection solution that train and simulate the existing of the attacker in every possible spot in the network diameter). The solution was tested in the same test-bed and yield about 79% overall accuracy. We build a misuseWireless Local Area Network Intrusion Detection System (WIDS) and discover some important fields in WLAN MAC-layer frame to differentiate the attackers from the legitimate devices. We tested several machine learning algorithms and found some promising ones to improve the accuracy and computation time on a public dataset. The best performing algorithms that we found are Extra Trees, Random Forests, and Bagging. We then used a majority voting technique to vote on these algorithms. Bagging classifier and our customized voting technique have good results (about 96.25 % and 96.32 %respectively) when tested on all the features. We also used a data mining technique based on Extra Trees ensemble method to find the most important features on AWID public dataset. After selecting the most 20 important features, Extra Trees and our voting technique are the best performing classifiers in term of accuracy (96.31 % and 96.32 % respectively).