Adversarial machine learning in Network Intrusion Detection Systems

Adversarial examples are inputs to a machine learning system intentionally crafted by an attacker fool the model into producing incorrect output. These have achieved great deal of success in several domains such as image recognition, speech recognition and spam detection. In this paper, we study nature adversarial problem Network Intrusion Detection Systems (NIDS). We focus on attack perspective, which includes techniques generate capable evading variety models. More specifically, explore use evolutionary computation (particle swarm optimization genetic algorithm) deep (generative networks) tools for example generation. To assess performance these algorithms NIDS, apply them two publicly available data sets, namely NSL-KDD UNSW-NB15, contrast baseline perturbation method: Monte Carlo simulation. The results show that our generation cause high misclassification rates eleven different models, along with voting classifier. Our work highlights vulnerability based NIDS face perturbation. • Machine not robust unconstrained domains. Evolutionary able successful examples. Generative Networks provide rich source fooling intrusion detection systems vulnerable maliciously packets.