Skip to content

Latest commit

 

History

History
62 lines (43 loc) · 3.06 KB

RESEARCH.md

File metadata and controls

62 lines (43 loc) · 3.06 KB
Raw

RESEARCH

This project is composed for a Deep Learning Model with Recurrent Neural Networks, with the objective of make easy the development of this part, this directory contains this elements:

  • NOTEBOOKS: For reply and create the model.Go to Notebooks

  • GENERATED MODELS: You can use differents models previosly generated.Go to Models

You can contribute with this project

You can create new models and upload to our models directory, also you can test with waf-benchmark.

alt

Thanks to the rooted mail list.

That helped us attacking our honeypots. Here are the promised results:

  • The first parameter is the percentage of sql injection attack.
  • The second parameter is the time to process this payload.
  • The third parameter is the payload.
  • The last parameter is the weight inside the network, the first element is the loss weight and the second is a binary element, if the network match this element. This is important for explicability, because you can detect patterns of attacks.

FINAL RESULTS

MODSECURITY

These are our results with waf-benchmark and this model

Modsecurity is based on regular expressions.

Tool name Attacks blocked Success attacks
sqlmap 20586 1876
OWASP ZAP 47952 9700
Payloads False Positives Passed
Darkweb 2017 Top 10000 0 20000
Family Names USA Top 1000 0 2000
Female Names USA Top 1000 0 2000
Male Names USATop 1000 0 2000
Names 0 20326

WAF BRAIN

Waf-brain is based on Deep Learning.

Tool name Attacks blocked Success attacks
sqlmap 21626 832
OWASP ZAP 49048 8206
Payloads False Positives Passed
Darkweb 2017 Top 10000 36 19872
Family Names USA Top 1000 0 2000
Female Names USA Top 1000 0 2000
Male Names USATop 1000 0 2000
Names 4 20322