Utilizing deep abstraction for higher intrusion-detection in wireless networks

Abstract

The recent advances in mobile technologies have resulted in IoT-enabled devices becoming more pervasive and integrated into our daily lives. The connected devices are ubiquitous, generating huge, high-dimensional and complex data. Observing malicious activities, which deviate from normal behavior, in such colossal data is a challenging task. Feature learning, however, can be one solution to solve this task. This dissertation thus proposes novel feature-learning schemes using Deep-Feature Extraction and Selection (D-FES) and fully unsupervised method. D-FES combines stacked feature extraction and weighted feature selection. The stacked auto-encoding is capable of providing abstractions (representations) that are more meaningful by reconstructing the relevant information from its raw inputs. The representations could also be leveraged as a clustering method. We then combine this with modified weighted feature selection inspired by an existing shallow-structured machine learning. We finally demonstrate the ability of the condensed set of features to reduce the bias of a machine learning model and the computational complexity of training and testing. We verify our proposed schemes on a Wi-Fi network dataset, called, the Aegean Wi-Fi Intrusion Dataset (AWID), prove the usefulness and the usability of the D-FES by achieving an impersonation detection accuracy of 99.918% and a false alarm rate of 0.012%. D-FES also achieved 99.910% of accuracy during all class of attacks detection.