Creating an Effective Intrusion Detection System Using a Deep Learning Algorithm

Started Sept. 1, 2022

Abstract or project description

Over the past few decades' internet traffic has grown exponentially, leading to an alarming rise in cyber attacks. Intrusion Detection Systems (IDS) provide security for networks by continuously monitoring activity and identifying malicious actors. Unfortunately, most of the IDS models employed today aren’t effective in real-world scenarios since they aren’t trained on real-world data, or they suffer from latency limitations due to their reliance on a wide breadth of input signals. As a result, a need for an effective IDS exists and this need will only grow with time. This research aims to provide a novel contribution to the field of cybersecurity by training an effective IDS, capable of being depl oyed in the real world. This model is trained on datasets created by gathering data on real networks (KYOTO 2006+ and UNSW-NB15) with genuine traffic, as opposed to being trained on simulated datasets, like most existing models. In addition, the model employs deep learning to help reduce the number of features used, making the model much more generalizable and computationally inexpensive. In order to properly benchmark the performance of this deep learning solution, baseline machine learning models were implemented and evaluated for comparison purposes. The final deep learning model achieves an impressive accuracy score of 99.42% on UNSW-NB15 and 99.54% on KYOTO 2006+, while only using 9 common features. This makes the final deep learning model extremely effective and generalizable, leading to significant potential for enhancing network security.