README.md

Autoencoder Feature Residuals for Network Intrusion Detection: Unsupervised Pre-training for Improved Performance

This repo contains the code needed to support the paper titled "Autoencoder Feature Residuals for Network Intrusion Detection: Unsupervised Pre-training for Improved Performance". Note that this code contains dependencies on external services such as Weights and Biases and other python libraries.

Extended book chapter code for recurrent neural networks

The code for the extended book chapter for recurrent neural networks can be found here: https://github.com/WickedElm/feature_residuals_with_pretraining_rnn

Executing Code

Assuming all of the dependencies are in place such as an account on Weights and Biases one can execute experiments by each dataset or all at once.

Running an experiment for all datasets

To do this simply clone the repo, change to its directory, and execute:

./run_all_tests

This will download the data needed and execute what we considered a single experiment in the paper.

Running an experiment for a specific dataset

To do this, clone the repo, change to its directory, and execute:

./download_data
./run_<dataset to run>

where you replace with one of the dataset run scripts in the repo such as run_nf_unsw_nb15_v2

Finding results data

After performing an experiment the resulting data can be found in the ./outputs/ directory.