ARDL-IDS: Adversarial Resilience in Deep Learning-based Intrusion Detection Systems

Abstract

With the growing complexity of computer networks and malicious attacks, countermeasures for prevention, detection, and protection are in high demand. Intrusion Detections Systems(IDS) have shown a lot of potential in detecting these attacks, but an effective and adaptive IDS that can be scaled and updated systematically is essential, and the efficiency of deep learning methods for the same has been rapidly increasing. ARDL-IDS identifies the need for a high-performing network intrusion detection system that sustains itself in the rapidly growing network environments and uses Deep Learning techniques to help in handling the volume and variety of these intrusions that are prevalent. The KDDCUP'99 dataset is used to classify the various types of attacks using a Deep Neural Network(DNN). To prevent possible attacks on the trained model, the system is made more robust through adversarial training with FGSM, BIM, MIM, and PGD attacks.