Reduce Low-Frequency Distributed Denial of Service Threats by Combining Deep and Active Learning

This study introduces substantial contributions to the field of Low-Rate Detection of Distribution Denial of Service (DDoS) Attacks, leveraging convolutional neural networks (CNNs) with an attention mechanism and incorporating active learning with semi-labelled data. These contributions collectively enhance the accuracy, efficiency, and adaptability of DDoS detection systems. Additionally, the study introduces active learning strategies into the learning process. By actively selecting instances for manual labelling, the study reduces the burden of extensive manual labelling efforts and enhances the model’s scalability. In the dynamic realm of cybersecurity, where threats evolve rapidly, active learning ensures the model’s adaptability with minimal human intervention. Furthermore, this research addresses the challenge of scarce labelled data in real-world cybersecurity contexts. By harnessing semi-labelled data efficiently and pairing it with active learning, the study streamlines the detection and defence against DDoS assaults with low attack rates. This is particularly relevant in situations where procuring an abundance of labelled data is impractical or cost-prohibitive.