Unsupervised network intrusion detection systems for zero-day fast-spreading network attacks and botnets

Today, the occurrence of zero-day and complex attacks in high-speed networks is increasingly common due to the high number vulnerabilities in the cyber world. As a result, intrusions become more sophisticated and fast to detrimental the networks and hosts. Due to these reasons real-time monitoring, processing and intrusion detection are now among the key features of NIDS. Traditional types of intrusion detection systems such as signature base IDS are not able detect intrusions with new and complex strategies. Now days, automatic traffic analysis and anomaly intrusion detection became more efficient in field of network security however they suffer from high number of false alarms. Among all type of anomaly detection methods unsupervised machine-learning techniques are commonly applied in NIDS to detect unknown and complex attacks in the network without any prior knowledge. This dissertation manly focuses on analyzing network traffic to find abnormal behavior in real time. The proposed framework consists of network traffic preprocessing, anomaly detection and clustering methods. The proposed framework is capable of generating meaningful reports related to the detection of real intrusions in well-known datasets. Unsupervised learning methods are capable of adapting their required features to the dynamically behavior of the network. Due to unfeasibility of payloads checking in high-speed network the proposed framework monitors network flows instead. Network flow contains the behavior of the network in higher extensive vision and shows the explicitness of the network data, which results in faster and higher detection rate of network attacks. This research shows that by using proper data preprocessing and unsupervised data analyzing methods it is possible to detect fast and complex zero days (new) attack in real time. The practical experiments are presented in the included articles