Deep Neural Network based Intrusion Detection System: A Comparative Analysis
Abstract - As more crucial infrastructure modernization and organizational digital transformation continue, the demand for improved and advanced cybersecurity measures and practices increases. Network intrusion detection systems (IDS) development is being pushed as a mean to mitigate cyber- attacks taking place in computer networks. Machine learning based IDSs are widely popular and are the subject of intense research and experimentation. Those traditional ML modes can be bypassed by modern attacks. Deep learning-based IDS, in contrast, does not require a lot of attack signatures to come up with an advanced solution. In this paper we design LSTM-based three deep models dubbed as LSTM30, LSTM60 and LSTM30+. We used CIC-IDS2017 which has been one of the most recent and up-to-date IDS datasets. This dataset consists of nine kinds of advanced attack samples. We explore various other traditional machine learning-based classifiers and models that exist in the literature and compared their performances with our models. The result of this study shows that LSTM30+ gives superior performances out of these three models with an accuracy of 98.62%.
Keywords - Intrusion Detection System, Machine Learning, Deep Learning, LSTM