An Autoencoder-based Efficient Scheme for DDoS Detection

In today’s time, as businesses and organizations rely heavily on their online presence to carry out their operations and most of the operations are being managed and performed by the different stakeholders being remote, Distributed Denial of Service (DDoS) attacks have become a significant concern for the industries. Most of the existing models for detecting DDoS attacks were evaluated using only a single dataset which does not represent the entire range of possible network traffic and attack scenarios. We propose an autoencoder-based approach for DDoS detection using a Double Autoencoder model, which combines the power of unsupervised learning with supervised learning. The proposed approach is evaluated on the CICDDoS2019 and NSL-KDD datasets and is able to outperform state-of-the-art DDoS detection techniques. The model uses the reconstructed error values to distinguish between normal and DDoS traffic and is able to learn both the normal and abnormal traffic patterns. The proposed method is also able to generalize well to different datasets, demonstrating its potential for real-world applications.