Comparative Study on Feature Selection Methods Rooted in Swarm Intelligence for Intrusion Detection

Today, the cyber-space landscape is filled with vulnerabilities and malicious users that exploit them. Adding to this the technological context, with data flooding from all directions, it is evident that maintaining a good level of security for our systems has become a challenge. The high-dimensional datasets generated by all these diverse technologies needs to be processed efficiently. Inside the pool of data records gathered from all the monitored devices, there is also noise that could negatively influence intrusion detection. Feature selection methods(FSM) may offer a plausible solution to this multi-dimensionality problem, because they reduce the number of features and keep only those significant for solving the stated problem. In this paper we propose to conduct a comparative study of feature selection methods for intrusion detection. We focus on wrapper variants of FSM which are based on swarm intelligence algorithms. To conduct our study, we construct our FSM models based on four SI algorithms (PSO, BA, BAL and BAE) in combination with traditional classifiers (SVM, C4.5 and Naive Bayes) and use the NSL-KDD dataset for our test and comparative analysis.