Host based intrusion detection using RBF neural networks

Abstract

A novel approach of host based intrusion detection is suggested in this paper that uses Radial basis Functions Neural Networks as profile containers. The system works by using system calls made by privileged UNIX processes and trains the neural network on its basis. An algorithm is proposed that prioritize the speed and efficiency of the training phase and also limits the false alarm rate. In the detection phase the algorithm provides implementation of window size to detect intrusions that are temporally located. Also a threshold is implemented that is altered on basis of the process behavior. The system is tested with attacks that target different intrusion scenarios. The result shows that the radial Basis Functions Neural Networks provide better detection rate and very low training time as compared to other soft computing methods. The robustness of the training phase is evident by low false alarm rate and high detection capability depicted by the application