A Hybrid Deep Learning Framework for Efficient Network Intrusion Detection Systems using Data Augmentation

Because of their critical position as the initial barrier against harmful actions, intrusion detection systems (IDSs) rely substantially on the efficacy of cybersecurity in protecting corporate communication. There are several limits to machine learning approaches that be solved by utilizing other deep learning architectures, even though they are widely used for intrusion detection. In addition, unbalanced datasets frequently make it difficult to evaluate the suggested models, which prevents a thorough evaluation of the models' effectiveness. This study seeks to tackle these difficulties by improving the performance of numerous deep-learning(DL) structures for IDS using data augmentation approaches. The experimental results shown that even a basic CNN-based architecture may produce very efficient network-attack-detection, but more complicated designs only managed to show slight performance increases. The results show that cybersecurity frameworks can easily incorporate the suggested deep learning-based intrusion detection approaches, which improves the capacity to identify and counteract complex network threats. The outcomes of this research exhibit that the accuracy of intrusion detection models is highly dependent on the dataset's quality and quantity, with augmented CIC-IDS-2017 showing an accuracy of up to 92%.