An Extensive Survey on Intrusion Detection Systems: Datasets and Challenges for Modern Scenario

Abstract

Cyberattacks are becoming more and more advanced, making it more difficult to identity suspicious activities on network traffic. Weaponizing the data in the line between network attacks and data breaches continues and the number rises upward even during the recent year with a massive increase in the attack type. Many consider Intrusion Detection System (IDS) datasets publicly available are becoming outdated and inadequate due to the availability of newer attack techniques. Therefore, it is a concern that the extensive usage of these available datasets in the current attack scenario to evaluate IDS models. This paper lists 37 datasets available for testing the IDS models and discusses those publicly accessible datasets, describing the characteristics and limitations for researchers who use such datasets. Finally, based on the dataset characteristics and usage survey, we conclude with a summary of the problems and provide our insights and suggestions for the use of network-based datasets for the Deep Learning approach for further improvement.