Intrusion detection using autoencoder

I was trying to implement a solution for detecting intrusions using an Autoencoder where a packet is either considered benign or attack (binary classification ) . I still find some problems in my notebook (For instance the cost that I get is very large) ,and I would like it if someone could help me in understanding the origin of the problem .
The code is available in this google colab link : Google Colab
And you may need to upload the KDDTrain+ dataset which you can find in kaggle : NSL-KDD (


Sounds interesting.
I’m going to move this to the “AI Projects” forum area.

1 Like

Thank you , can you give it a look .
I started this project last year but did not complete it , so I forgot why I implemented some stuff .
Now I think I wanted to train an autoencoder on some normal benign packets , and then set a threshold on the output of the autoencoder when predicting a certain packet : If the output of the autoencoder is less than the threshold , then the packet is benign , but if the output is higher , then we can predict that it is an attack .

1 Like

Sorry, I’m not able to investigate this further.