Most of the papers have compared different hyperparameters such as layers, initializers, activation function, and optimizer algorithm in the CNN algorithm for detecting network intrusion. This is practical, while in the scientific view, I have not found any proofs for it.
Network data itself is not a kind of image data and is converted to an image through the process.
How can we prove in scientific view that using CNN algorithm not only properly analyzes images (with multiple dimension data like pixel, domain, phase, and color), but also well-performed in a network data (with just one dimension: features) converted to an image (features depict as pixels)?
Maryam Hatami
Please see if you get any help from the article below.
https://www.researchgate.net/publication/341555600_CNID_Research_of_Network_Intrusion_Detection_Based_on_Convolutional_Neural_Network
Thank you very much, but I needed a deeper scientific reason. Showing forward propagation, error, and backward propagation for data in multi-dimension vs one-dimension.
We have very good articles for this query.
Plz see
Article Network intrusion detection system: A systematic study of ma...
Thank you very much, but I meant mathematical proofs for changing network data dimension in CNN and not evaluation metrics :)
- Badges
- Science topic