Border Router Intrusion Detection System. It's not birds, btw!
An intrusion detection system using network-flow statistics collected at a border router. A minor project done for Advanced Computer Networks course at Department of IT, NITK.
To access the performance of the 1D-CNN, and Random Forest classifier on unormalized and normalized network-flow statistics, with and without autoencoder.
- Used
dvc(https://dvc.org/) for pipeline construction) - Tensorflow 2 with
tf.kerasfor Neural network scikit-learnfor Random Forest
- Random Forest works well both with normalized and unnormalized data
- Neural networks needs normalized data for better performance
- Auto-encoders do a better job in encoding unnormalized data, as evident by the performace of
Unnormalized+Auto-encoder+Random Forestconfiguration
The methods and results were supposed to be published at a conference, but was rejected as the paper was poorly written and conclusions were not significant. It was communicated to another conference as-is, which got accepted with a suggestion to make the title short š¤£; which was then dropped by me. The pdf of the paper is provided, just for reference. It is not to be taken seriously.