An Efficient DDoS Attack Detection Scheme using Autoencoder based on Feature Selection Approaches

Distributed Denial-of-Service (DDoS) attacks is an important threat to the security and availability of network services. To address this encounter, we propose an efficient DDoS attack detection scheme that leverages an autoencoder (AE) approach is combined with feature selection using the XGBoost algorithm. The technique begins by applying the XGBoost feature importance mechanism to choose the most applicable features from the dataset, thereby reducing dimensionality and computational overhead. An autoencoder is then trained on the selected features to learn a compact and healthy representation of the network traffic. By comparing the reconstruction error of the autoencoder against a predefined threshold, the system efficiently distinguishes between malicious traffic and normal. The proposed approach attains high detection accuracy and real-time performance, making it appropriate for deployment in modern networks. Experiments conducted on benchmark datasets prove the efficacy of the scheme in detecting a wide range of DDoS attack patterns while maintaining computational efficiency.