SCDAE-MLP Design for Detecting and Classifying DDoS Attack

This paper addresses the dataset imbalance problem with enhancing the accurate detection capability of distributed denial of service (DDoS) attacks. To generate balanced datasets, self-training mixup decision tree (STM-DT) with random sampling is introduced. A novel cybersecurity algorithm named SCDAE-MLP is designed, where stacked convolutional denoising autoencoder (SCDAE) is used to capture the latent feature of network traffic and multilayer perceptron (MLP) is served as a classifier who effectively distinguishes between normal and DDoS attack traffic. Experimental results demonstrate that SCDAE-MLP excels in automatically extracting low-dimensional, high-quality features while demonstrating strong generalization capabilities to handle large-scale and highly variable attack data.