The increase of security threats in computer networks requires the implementation of several lines of defense capable to effectively preserve the confidentiality, integrity and availability of the information. In the present work multilayer perceptron (MLP) artificial neural networks are trained to be used for monitoring events and identifying possible attempts to compromise the network resources. The 1999 DARPA dataset was taken as source of traffic packets. Specific packet information including some relevant features of the application layer and of the TCP/IP protocols headers were selected for training and testing these neural networks. As a result, several neural networks were tested and their optimal design parameters were derived to ensure an efficient detection of anomalies in the network traffic. The feasibility of using MLP networks for intrusion detection was confirmed in such a realistic case.
Conference: 13 Conferência sobre Redes de Computadores (CRC) in Leiria. Portugal