A GA-LR wrapper approach for feature selection in network intrusion detection

Abstract

Intrusions constitute one of the main issues in computer network security. Through malicious actions, hackers can have unauthorised access that compromises the integrity, the confidentiality, and the availability of resources or services. Intrusion detection systems (IDSs) have been developed to monitor and filter network activities by identifying attacks and alerting network administrators. Different IDS approaches have emerged using data mining, machine learning, statistical analysis, and artificial intelligence techniques such as genetic algorithms, artificial neural networks, fuzzy logic, swarm intelligence, etc. Due to the high dimensionality of the exchanged data, applying those techniques will be extremely time consuming. Feature selection is needed to select the optimal subset of features that represents the entire dataset to increase the accuracy and the classification performance of the IDS. In this work, we apply a wrapper approach based on a genetic algorithm as a search strategy and logistic regression as a learning algorithm for network intrusion detection systems to select the best subset of features. The experiment will be conducted on the KDD99 dataset and the UNSW-NB15 dataset. Three different decision tree classifiers are used to measure the performance of the selected subsets of features. The obtained results are compared with other feature selection approaches to verify the efficiency of our proposed approach.