Optimization of Network Intrusion Detection System Using Genetic Algorithm with Improved Feature Selection Technique

The on-line environment is growing in unimaginable speed and scale. It offers convenience in our professional and personal life but on the other hand, exposes us to threats and danger in terms of network intrusion that eventually leads to invasion of privacy and other network security issues. This paper has proposed an optimization of the Network Intrusion Detection System (NIDS) using Genetic Algorithm with improved feature selection (GA-IFS) technique. GA-IFS monitors and analyzes network simulated activities using DARPA KDD Cup 99 dataset to efficiently identify normal and anomalous network traffic. GA is a search heuristic that is suitable for problems with large population size, however, it has drawbacks with regards to time taken for convergence which sometimes leads to local optima if lack of population diversity. This became an opportunity to enhance GA by integrating Support Vector Machine (SVM) classifier for selecting feature subset, reducing dataset dimensionality, identifying relevant features and improving the intrusion detection rate. Experimental results show the significant effect of dataset preprocessing procedure of GA-IFS in removing redundant and irrelevant records of 79.07% training data and 80.47% test data. With the implementation of improved feature selection, R2L got the highest improvement of 3.87% detection rate. This is followed by DOS with 3.33%, Normal with 2.76%, Probe with 1.22% and U2R with 0.91% compared to traditional GA.