Mitigating the Knock-on-Effect of DDoS Attacks on Application Layer using Deep Learning Multi-Layer Perception

Abstract

Distributed Denial of Service (DDoS) attacks are a significant threat to the Internet today, and application-layer DDoS attacks that use legitimate HTTP requests to overwhelm victim resources are imperceptible. As a result, neither the intrusion detection system (IDS) nor the victim's server can detect malicious packets. This research is intended to take a new approach, deep learning, to detect novel application layer DDoS attack strategies, known as increasing DDoS and proxy DDoS attack. The proposed attack detection model has been validated by performing simulation experiments with MATLAB and Python. Finally, the preliminary result compared with the various machine learning techniques for classification: Naïve Bayes, SVM, Decision trees, and Genetic Algorithm (GA) in terms of Accuracy Rate (AR), Detection Rate (DR), Sensitivity, specificity, (ROC) curve. The results show that our detection model effectively detects DDoS attacks at the application layer. The proposed deep learning multi-layer perceptron architecture can identify and use the most relevant high-layer features of packet flows with an accuracy of 98% on the generated dataset containing a novel DDoS attack.