Behavior-based outlier detection for network access control systems

Network Access Control (NAC) systems manage the access of new devices into enterprise networks to prevent unauthorised devices from attacking network services. The main difficulty with this approach is that NAC cannot detect abnormal behaviour of devices connected to an enterprise network. These abnormal devices can be detected using outlier detection techniques. Existing outlier detection techniques focus on specific application domains such as fraud, event or system health monitoring. In this paper, we review attacks on Bring Your Own Device (BYOD) enterprise networks as well as existing clustering-based outlier detection algorithms along with their limitations. Importantly, existing techniques can detect outliers, but cannot detect where or which device is causing the abnormal behaviour. We develop a novel behaviour-based outlier detection technique which detects abnormal behaviour according to a device type profile. Based on data analysis with K-means clustering, we build device type profiles using Clustering-based Multivariate Gaussian Outlier Score (CMGOS) and filter out abnormal devices from the device type profile. The experimental results show the applicability of our approach as we can obtain a device type profile for five dell-netbooks, three iPads, two iPhone 3G, two iPhones 4G and Nokia Phones and detect outlying devices within the device type profile.