A comparative analysis of different soft computing techniques for intrusion detection system

Abstract

In this internet era, the data are flooded with malicious activities. The role of soft computing techniques to classify highly vulnerable, complex and uncertain network data by devising an intrusion detection system is so significant. The proposed work emphasizes on the classification of normal and anomaly packets in the networks by carrying out the comparative performance evaluation of different soft computing tools including Genetic Programming (GP), Fuzzy logic, Artificial neural network (ANN) and Probabilistic model with Clustering methods using NSL-KDD dataset. Here, Fuzzy logic runs the first place in the performance metrics and the clustering algorithms and Genetic programming deliver the worst performances. Fuzzy Unordered Rule Induction Algorithm (FURIA) in Fuzzy logic gives a high detection rate of accuracy (99.69%) with the low rate of false alarms (0.31%). The computational time of FURIA (78.14 s) is not so expectant. So Fuzzy Rough Nearest Neighbor(FRNN) is recommended as an optimistic model with a sensible accuracy rate of 99.51% and tolerable false alarm rate of 0.49% along with a pretty good computational time of 0.33 s.