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

•First, the preprocessing stage consists of resampling, changing the attribute values, and removing redundant records.•Second, the feature selection stage reduces the feature space of the used datasets using the GA-LR wrapper.•Third, the classification stage is performed using three decision tree classifiers namely with the best selected subsets.•The best subset for the KDD99 is composed of 18 features and gives an accuracy equal to 99.90%, 99.81% DR and 0.105% FAR.•The best subset for the UNSW-NB15 is composed of 20 features and gives an accuracy equal to 81.42% and 6.39% FAR.