Intrusion detection systems (IDS) are systems that are capable of monitoring networks for malicious events and policy violations. They are systems that are capable of detecting and classifying network attacks based on behaviors or signatures of known attacks. However, since network attacks are constantly evolving the effectiveness of IDS is often challenged as a result of novel attacks generated, known as Zero-day attacks. Recently, the Machine Learning (ML) based technique used in identifying malicious attacks, uses a supervised-based learning approach that is unable to classify anomalies rendering it ineffective because Zero-day attacks do not exist in the training data set. However, unsupervised learning systems are capable of learning what is usual for a particular data set and then detecting the differences in new unclassified data. The constant change in network behavior and the failure of ML-based IDS to detect Zero-day attacks facilitate the need to evaluate Deep Learning (DL) algorithms that can effectively detect Zero-day attacks in Intrusion Detection Systems. This paper provides a comprehensive survey and evaluation of different Deep Learning algorithms in order to further propose the implementation of DL-based IDS.
Authors: Sunkanmi Oluwadare, Dr. Zag ElSayed