الفهرس | Only 14 pages are availabe for public view |
Abstract In recent years, the networking systems have witnessed a breakthrough. Due to the advances in network and communication technology, new concepts such as Internet of Things (IoT) have let network configurations to emerge. The IoT framework connects devices such as computers, mobile, etc. The nature of IoT networks makes them vulnerable to attacks introduced by malicious insiders, denial of services, and web-based attacks. These attacks may introduce worms or viruses into the IoT network. Software-Defined Networking (SDN) technology has been extended and implemented in IoT networks. This technology reduces the hardware cost and allows separation of data and control planes of traffic. SDN can be merged with IoT because SDN provides opportunities to solve issues related to IoT security. Designing a centralized SDN controller is another challenge as there is a need to monitor and implement real-time intrusion detection in high-speed networks. To deal with the complicated scenarios, there should be tools that can detect attacks as early as possible. This can be done by Intrusion Detection Systems (IDSs). The idea of operation of these systems depends on anomaly detection tools and concepts. These systems should be built upon trained classifiers. Both Machine Learning (ML) and Deep Learning (DL) tools can be used to achieve this task. Both ML and DL tools work on signals like data acquired through data or control planes generated in the SDN architecture. The utilization of SDN leads to high quality of the IDS performance. In this thesis, an efficient IDS based on Convolutional Neural Network (CNN) for binary classification is introduced and applied on new attack-specific SDN dataset. The simulation results showed that the CNN model achieved the highest accuracy of 99.2 %. In addition, the effectiveness of hybrid (Long Short Term Memory) LSTM-CNN for binary classification is analyzed for IDS and applied on a new attack-specific SDN dataset. The simulation results show that the hybrid CNN-LSTM model achieved the highest accuracy of 99.8 %. Another deep CNN model for multi-class attack has been proposed and it achieved the highest accuracy of 99.85 %. |