Unveiling the Effectiveness of CNN-based Models for Multiclass DDoS Attack Detection and Classification: A Comparative Analysis

The network infrastructure faces increasing vulnerability to cyber-attacks due to the growing array of services available over the Internet. Distributed Denial of Service (DDoS) attacks are among the most disruptive attacks, which fill servers and networks with excessive traffic and overload websites and online services. This research delves into classifying diverse DDoS attack types using powerful Convolutional Neural Network (CNN) models. Building upon prior research distinguishing attack and non-attack instances, our investigation intricately explores the identification and categorization of multiclass DDoS attacks. The study undertook three meticulous experiments with the dataset CIC DoS 2019 with 68 columns encompassing 12 attack and non-attack classes. The first involved utilizing the complete dataset while accounting for class imbalance. Subsequently, a balanced dataset was constructed for the second experiment, followed by a thoughtful dataset reduction of 60Three highly distinguished CNN models were introduced for the classification task. A simple yet informative CNN Model that serves as the baseline for benchmarking, a Complex CNN Network that embodies the capability for profound feature learning. Additionally, a Complex CNN Network fortified with Regularizers demonstrates an unwavering dedication to improving generalization amidst complexity. Our comprehensive analysis focuses on four crucial performance parameters: Accuracy, Loss, Precision, and Recall. The results unequivocally endorse the simple neural network as the pinnacle of efficacy. This model showcases remarkable accuracy, minimal loss, and exceptional precision and recall in identifying DDoS attacks amidst the data deluge.