Honeypot is a decoy system with vulnerabilities introduced to trap hackers. Through many years of evolution, a generation of smart honeypots has been developed. The self-adaptive honeypot is a smart honeypot that is expected to respond appropriately to the attacker's request. In most existing self-adaptive honeypot systems, the commands sent from the attacker play a central role in the reasoning process of the honeypot. In this thesis, we focus on the construction of a machine learning workflow that aims at estimating the risk level of these commands. Experiments show that the proposed workflow achieves potential results.
|Date of Award||4 Sep 2020|
|Supervisor||Jean-Noel COLIN (Supervisor)|
- machine learning
- representation learning
- labeling model