Can anyone recommend some good sources of annotated (labeled) datasets for network security tests and Machine Learning (ML)? In general, various cybersecurity areas are welcomed but from reliable and confident sources. Poor and incorrect annotations or malicious sources are not of interest, so avoid it please.
Dear Dragan Pleskonjic,
I found this GitHub repo, where there is a list of CyberSecurity datasets:
"https://github.com/jivoi/awesome-ml-for-cybersecurity#-datasets". In addition, these links can be useful:
Conference Paper A Review on Cyber Security Datasets for Machine Learning Algorithms
Hello,
unforutunately real-world annotated cybersecurity datasets are rare. Actually, lack of such datasets enforces somehow the adoption of unsupervised approaches to event detection.
Take a look at the following paper if your are interested:
Empirical analysis and validation of security alerts filtering techniques, IEEE transaction on Secure and Dependable Computing.
Best,
Andrea
Andrea Paudice - thank you very much for your answer and pointing me to your very interesting paper.
Balachandra Muniyal - Thank you. We are aware of NSL-KDD datasets.
Did you tried CICIDS2017 dataset?
www.unb.ca/cic/datasets/ids-2017.html
We recently released the OmniDroid dataset, containing a set of static and dynamic features extracted from Android benign and malicious samples.
The link is below. Hope you find it interesting.
https://aida.ii.uam.es/datasets/
While neither huge in size nor specific to network security, Ghera is a repository of 50+ vulnerability benchmarks based on known Android app vulnerabilities.
- https://secure-it-i.bitbucket.io/ghera/index.html
- https://www.researchgate.net/project/Ghera-A-Repository-of-Android-App-Vulnerability-Benchmarks
One big advantage of Ghera is the benchmarks are authentic and demonstrable (due to duality) -- they contain vulnerable apps along with the malicious apps that exploit the vulnerabilities. Recently, we used Ghera to evaluate the effectiveness of Android security analysis tools. Preprint Are Free Android App Security Analysis Tools Effective in De...
- https://secure-it-i.bitbucket.io/rekha/index.html
- https://secure-it-i.bitbucket.io/rekha/dashboard.html
From ML perspective, I believe duality and labelled-ness may prove to be quite useful.
Currently, Ghera repo is closed to public due to maintenance. However, if you are interested in using the repo, then email us and we will try to assist you the best we can.
You can check the UNSW-NB15.
Here: https://www.researchgate.net/publication/320900660_A_new_evolutionary_neural_networks_based_on_intrusion_detection_systems_using_multiverse_optimization?_sg=ztD1Md4FQXu1RCbBPsYuA9WhSmzJkajutZI1DP7RHzjliOj5c7QzyoEXz7vgeyTH4eVEFJ_kQwGLTTF6o1EpwMHJ9_zgL_lVOxeeI4M8.w_R4EihpTWP60idxvY9vBQaWBWz_U7a_dIX69xJUEwvVh-F_9jHKcXtryV_IJ48EAUc59LljquIn6cbOToXZuA
Hello;
I suggest to use Kyoto 2006+ University Benchmark Dataset version 2015 , which is one of the most recently used intrusion detection dataset in the IDS domain. This dataset was developed by to solve the problems of underperformance of machine learning based IDS model trained and tested on KDD or NSL-KDD and any other related old datasets heavily criticized for not reflecting the current trend of network situation and sophistication of ever evolving cyber-attacks. Kyoto dataset is a multivariate attributes dataset that consists of 24 statistical features extracted from captured data including 14 conventional features that were derived from KDD Cup 99 benchmark dataset, and 10 additional features added for analysis and further investigation.
This dataset is available at this link :
http://www.takakura.com/Kyoto_data/new_data201704/
Dragan Pleskonjic : you could find a number of interesting dataset in the following link:
https://github.com/shramos/Awesome-Cybersecurity-Datasets
CIDDS dataset :
https://www.unb.ca/cic/datasets/ids-2017.html
The UNSW-NB15 Dataset :
https://www.unsw.adfa.edu.au/unsw-canberra-cyber/cybersecurity/ADFA-NB15-Datasets/
- Badges
- Science topic