Machine Learning Techniques for Characterizing IEEE 802. 11b Encrypted Data Streams

As wireless networks become an increasingly common part of the infrastructure in industrialized nations, the vulnerabilities of this technology need to be evaluated. Even though there have been major advancements in encryption technology, security protocols and packet header obfuscation techniques, other distinguishing characteristics do exist in wireless network traffic. These characteristics include packet size, signal strength, channel utilization and others. Using these characteristics, windows of size 11, 31, and 51 packets are collected and machine learning (ML) techniques are trained to classify applications accessing the 802.11b wireless channel. The four applications used for this study included E-Mail, FTP, HTTP, and Print. Using neural networks and decision trees, the overall success (correct identification of applications) of the ML systems ranged from a low average of 65.8% for neural networks to a high of 85.9% for decision trees. These averages are a result of all classification attempts including the case where only one application is accessing the medium and also the unique combinations of two and three different applications.Air Force Inst of Tech Wright-Patterson AFB OH School of Engineering and Management is the author of 'Machine Learning Techniques for Characterizing IEEE 802. 11b Encrypted Data Streams', published 2004 under ISBN 9781423519911 and ISBN 1423519914.