Published December 12, 2019 | Version 1

Machine Learning Datasets for Cyber Security Applications

  • 1. Military Technical Academy Ferdinand I, Romania

Description

Abstract: The main objective of this study is not to identify the best machine learning model, but instead to review the main datasets, publicly available, used to train and test security solutions that employ modern classification algorithms for anomaly detection. Hence, DARPA 1998 and KDD were studied as they were the first initiatives taken in this direction, while NSL-KDD, ISCXIDS2012 and CICIDS2017 are taken in consideration for future research because of their advantages. Personalized datasets will always bring a reasonable amount of uncertainty, especially since some feature vectors used for training remain unknown. Nevertheless, training on data specific to the protected infrastructure is more efficient, from the security point of view, than training on old attack signatures.

Files

2019mlds-2.pdf

Files (367.2 kB)

Name Size
md5:819cb455f13b98c4103e9264ee4acfe7
367.2 kB Preview Download