"Accurate intrusion detection technique based on deep learning for Software-Defined Networking (SDN)"

In recent years, Software Defined Networking (SDN) has gained traction as an innovative approach to overcome the limitations of legacy systems. The SDN's main goal is to separate the control and data planes, making network management easier and enabling for more efficient programmability. The suggested approach would protect the SDN in a more accurate, effective, and scalable manner, overcoming current methods' shortcomings. Different businesses employ network detection algorithms to arrange and differentiate harmful traffic these days, however there may be many challenges to encounter and determine assaults in imbalanced datasets. Examining the influence of different contemporary Deep Learning (DL) approaches, such as the Long Short-Term Memory (LSTM) based autoencoder and the Recurrent Neural Network (RNN), on the overall system functioning would increase the productivity of the anomaly-based Intrusion Detection System (IDS). DL is a relatively new topic of data security that is widely regarded as one of the most important solutions for addressing shortcomings in standard Machine Learning (ML) approaches. Deep learning can learn the nonlinear structure of data with a large number of dimensions. As a result of its ability to automatically identify connections in raw data without the need for human interference, it can increase the intrusion detection rate. DDoSNet, an intrusion detection technique for Distributed Denial of Service (DDoS) assaults in SDN circumstances, is proposed in this thesis. It will improve the performance of the anomaly-based intrusion detection system by examining the influence of several existing DL techniques, such as the basic RNN-based autoencoder and the LSTM-based autoencoder, on the overall system functioning. We tested our model using the InSDN and CICIDDoS2019 datasets, both of which were recently published. Because our major goal is to tackle the problem of binary classification in the Intrusion Detection System, we compare both datasets and their outcomes in our suggested approach (IDS). We use the InSDN dataset, which is an attack-specific SDN dataset, published in the year 2020. The benign and various attack categories that may occur in the various elements of the SDN platform are included in this new dataset. While CICIDDoS2019 is a collection of benign and up-to-date popular DDOS attacks that closely resembles real-world data. This dataset includes a broad range of Distributed Denial of Service attacks. As compared to different benchmarking techniques of InSDN dataset, in CICIDDoS2019 a significant enhancement was obtained in the context of attack detection. As a result, our approach gives us a lot of confidence when it comes to securing these networks.