Optimized Multi - Layer Hierarchical Network Intrusion Detection System with Genetic Algorithms

The number of connected devices on the Internet has exceeded 31 billion devices in 2018 and it is forecasted that this number will exceed 50 billion by the year 2020. One the other hand, malicious software and network attacks are raising on an alarming rate. It is estimated that more than 230,000 new malware are produced daily and over 53,000 new Cryptoware malware engines are detected as well. This proliferation in security attacks constitutes a great challenge for Intrusion Detection Systems (IDS), in particular, in detecting modern attacks. In this paper, a multi-layer hierarchical Network Intrusion Detection System (NIDS) is proposed with the aim to improve the overall detection performance of the IDS for detecting modern attack types. The proposed multi-layer NIDS utilizes multiple models of machine learning algorithms in a hierarchical architecture in addition to using evolutionary computing, namely Genetic Algorithms, to tune the configurations of the neural network models used in the first layer. A modern dataset (i.e. CICIDS-2017) is used to evaluate the proposed approach, which contains several modern attacks. The results showed that the proposed multi-layer system significantly improved on the error generalization metrics.