Classification of Network Intrusion Detection System Using Deep Learning

Over the past  one decade, there has been a continuous rise in the usage of Internet services all over the world. However, numerous challenges emerge since malicious attacks are constantly changing and are happening in exceptionally huge volumes requiring an adaptable solution. This has led to a desperate need not only of detection and classification of attacks at host as well as network side but also the detection being automatic and in a certain time frame, as a result of which the world has seen many developments in this field with machine learning and deep learning playing a huge role in it. Because of the dynamic effect of malware with constantly changing attack techniques, the malware datasets accessible openly are to be updated efficiently and benchmarked. In order to develop an effective intrusion detection system, machine learning or deep learning techniques are also becoming more advanced day by day, and it is important to utilize their benefits in this field. This paper focuses on the development of network intrusion detection systems (NIDS) using deep learning. This paper uses UNSW-NB15 dataset as it is one of the most recent and improved IDS datasets. It has been improved on many factors from its predecessor KDD CUP99. Convolutional neural network and recurrent neural network have been implemented to compare the results. The classifications implemented in this paper are both in binary and multiclass with the major focus regarding maximum macro precision, recall, and f-score for the multiclass approach.