ZED IDS prototype

The code and data files that can be downloaded here allow to reproduce the experiments on the use of a single autoencoder
for intrusion detections documented in the papers:

"Simpler Is Better: On the Use of Autoencoders for Intrusion Detection"
(Marta Catillo, Antonio Pecchia, Umberto Villano), In Quality of Information and Communications Technology, 2022
https://link.springer.com/chapter/10.1007/978-3-031-14179-9_15
http://ultraviolet.ding.unisannio.it/villano/papers/QUATIC2022_prereview.pdf (pre-review manuscript)

and

"Intrusion Detection with a Single Deep Autoencoder: Theory and Practice"
(Marta Catillo, Antonio Pecchia, Umberto Villano)
currently under review for publication on Software Quality Journal

Credits should be given to the developers by referencing the above papers.


The TRAIN, VALIDATION and TEST data files provided here were extracted from the modified CICIDS2017
dataset published at https://intrusion-detection.distrinet-research.be/WTMC2021/tools_datasets.html

Download ZED_distro_SQJ2022.zip

IMPORTANT NOTE:
Please do not be disappointed with the performance results obtained in standard mode.
As discussed in our papers, the quality of the autoencoder depends on seed and hardware used
for training
.

The sample pre-trained autoencoder supplied with the software (AE.tgz) is the best AE we found for 77
features on the CICIDS2017 dataset. Running it in test mode (see manual for details) makes it possible
to obtain the performance figures shown in our papers.


Made with ‌

Mobirise.com