A Competitive Neural Network for Intrusion Detection Systems

Detecting network intrusions is becoming crucial in computer networks. In this paper, an Intrusion Detection System based on a competitive learning neural network is presented. Most of the related works use the self-organizing map (SOM) to implement an IDS. However, the competitive neural network has less complexity and it is faster than the SOM, achieving similar results. In order to improve these results, we have used a repulsion method among neurons to avoid overlapping. Moreover, we have taken into account the presence of quantitative data in the input data, and they have been pre-processed appropriately to be supplied to the neural network. Therefore, the current metric based on Euclidean distance to compare two vectors can be used. The experimental results were obtained by applying the KDD Cup 1999 benchmark data set, which contains a great variety of simulated networks attacks. Comparison with other related works is provided.