حسن اسدي رباطتركي

Software-Defined Networking (SDN), which is a symmetric, dynamic and available network architecture, is proposed as a solution to create a more flexible network infrastructure with programmable devices. Using SDN, the entire system controller can be controlled using a centralized remote device. The benefits of SDN have led many companies to deploy it in their network environments. On the other hand, intrusion detection system (IDS) is used to detect the presence of malicious or unwanted packets in the network. This system works in real-time using traffic monitoring to determine if there is any unusual behavior on the network. Since SDN involves centralized management of the network by the controller, it aims to separate the functions of network control and data transmission. This makes network control and management more flexible. However, the network traffic data in the controller is the main target of hacker attack and also an important indicator for an intrusion detection system to identify malicious traffic. Integrating IDS into SDN is potentially one of the most reliable ways to secure these types of networks. In addition, it can integrate threat detection approaches into SDN through SDN characteristics, such as its programmability. For example, a potential factor related to intrusion detection is the packet statistics in a flow in the data page and the forwarding of the extracted field depending on the flow tables. Therefore, SDN can use these features for intrusion detection. In this thesis, a deep neural network approach to detect anomalies in network traffic is presented. The existing SDN dataset was used to eva‎luate the proposed method. First, pre-processing was performed on the data, then PCA dimensionality reduction was performed to achieve better results. By comparing the proposed method with two simple Bayesian and logistic regression algorithms, it was observed that the proposed method achieved a high accuracy rate of up to 99.4%, which is a good improvement compared to the other two classifications. Also, this method was compared with previous works. The obtained results show that the proposed method is optimal for implementation in the environment of SDN networks. Keywords: Network intrusion detection system, software-based networks, deep neural network, deep learning

سيستم تشخيص نفوذ شبكه (IDS) , شبكه‌هاي مبتني بر نرم‌افزار (SDN) , شبكه عصبي عميق , يادگيري عميق

Network intrusion detection system , software-based networks , deep neural network , deep learning

Hassan Asadi Robat Torki

Mohammad Abdollahi Azgomi