Abstract
Today, among the problems of computer security there is the one that the traditional Network Intrusion Detection System detect only a subset of the network traffic generated by attacks, that is the one that corresponds to the signature of known attacks.
In order to minimize this problem, this thesis proposes an innovative system based on neural networks to exhaustively identify the traffic generated by attacks by classifying flows.
For the purpose of training neural networks, dedicated systems for labeling, analysis and pre-processing of flows have been implemented.
An in-depth experimental campaign was carried out aimed at identifying the configuration of the system capable of delivering the best classification performance.
The final results were obtained by training and testing the best configuration over 24 hours of flow collected by the equipment of a broadband network.
|