On Classification of Alarms from Network Intrusion Detection System using Multi-layer Feed-forward Neural Networks

K. Goto and K. Keeni (Japan)


Intrusion Detection System, Data mining, Expert system, Back propagation


This study focuses on the subject of classification of alarm messages given by the Intrusion Detection System (IDS). A multilayer feed-forward neural network is employed for deciding the importance of the message given by the IDS. The messages given by the IDS have been collected and they are labeled by an expert for the training purpose. The original message fields have been normalized so that they can be used as inputs for the neural network. The training set has been partitioned into two parts, and 20 different ini tial seeds are used along with different number of hidden units for training the proposed neural network. The results obtained in the preliminary part of this work are promising. The average accuracy rate obtained by the network is 86.09 % with 8 hidden units.

Important Links:

Go Back