Analysis and Detection of DDoS Backscatter Using NetFlow Data, Hyperband-Optimised Deep Learning and Explainability Techniques

Abstract

The Denial of Service attacks are one of the most common attacks used to disrupt the services of public institutions. The criminal act of exhausting a network resource with the intent to obstruct the utility of a service is associated with hacktivism, blackmailing and extortion attempts. Intrusion Prevention Systems are an essential line of defence against this problem, strengthening public institutions, industrial and critical infrastructure alike. In the following work, an analysis of the detection of DDoS Backscatter with the use of neural networks is performed. To this end, a novel dataset is collected and described, on which a hyperband-optimized neural network is trained, and the decision process of the classifier is explained using LIME and SHAP.