Intrusion Detection and Attack Classification Using Feed-Forward Neural Network

Fast Internet growth and increase in number of users make network security essential in recent decades. Lately one of the most hot research topics in network security is intrusion detection systems (IDSs) which try to keep security at the highest level. This paper addresses a IDS using a 2-layered feed-forward neural network. In training phase, “early stopping” strategy is used to overcome the “over-fitting” problem in neural networks. The proposed system is evaluated by DARPA dataset. The connections selected from DARPA is preprocessed and feature range is converted into [-1, 1]. These modifications affect final detection results notably. Experimental results show that the system, with simplicity in comparison with similar cases, has suitable performance with high precision.