Key points are not available for this paper at this time.
Abstract By enabling the control and administration of the entire network from a single location, a Software-Defined Network (SDN) was created to streamline network administration. SDN controllers find intruders appealing because they make good targets. Attackers can take control of an SDN controller and use it to route traffic according to their requirements, which can have disastrous effects on the network. Although integrating SDN with deep learning strategies opens up novel avenues for IDS deployment defense, the detection models' efficacy depends on the quality of the training data. While deep learning for non-identifiable detection systems (NIDSs) has yielded promising results recently for several problems, most studies overlooked the impact of imbalanced and redundant datasets. Therefore, to improve the detection of network intrusions via binary and multiclass categorization, we proposed a novel enhanced ensemble DL-based Dual Parallel Attention Transformer (DPAT) with a Modular Deep Fully Convolutional Network (MDFCN), termed DPATMFNet approach. An Enhanced AlexNet method extracts the features from the input data. The Boosted Binary Meerkat Optimization Algorithm (BBMOA) is applied to choose the key features. The proposed system categorizes attacks, separates malicious from benign traffic, and identifies outstanding performance sub-attack types. Three of the most current realistic datasets were used for training and evaluation to demonstrate the effectiveness of the suggested system. We examined and contrasted its performance with that of other IDSs. The experimental findings indicate that the proposed system performs better than others at identifying various attacks. The suggested datasets achieve accuracy, detection rate, and precision above 99% compared to existing approaches. The results show how effective the proposed model is at obtaining high accuracy while requiring a shorter training period.
Saikam et al. (Mon,) studied this question.