Application of machine learning techniques for detecting anomalies in communication networks

Detecting, analyzing, and defending against cyber threats is an important topic in cyber security. Applying machine learning techniques to detect such threats has received considerable attention in research literature. Anomalies of Border Gateway Protocol (BGP) affect network operations and their detection is of interest to researchers and practitioners. In this Thesis, we describe main properties of the BGP and datasets that contain BGP records collected from various public and private domain repositories such as Réseaux IP Européens (RIPE) and BCNET. With the development of fast computing platforms, the neural network-based algorithms have proved useful in detecting BGP anomalies. We apply the Long Short-Term Memory machine learning technique for classification of known network anomalies. The models are trained and tested on various collected datasets. Various classification techniques and approaches are compared based on accuracy and F-Score.