Concurrent with the ever-increasing growth of information and communication technology (ICT) and the dramatic expansion ofcomputer networks, we observe different forms of attacks and intrusions to networks; thus intrusion detection systems (IDSs) are consideredas a vital part of each network connected to internet in the modern world. Neural networks are considered as a popular method used in IDS.Two major problems in these networks, i.e. long training time and inattention to features' domain, have made necessary development and/orimprovement of the model. Feature selection techniques are used in the neural networks in order to develop a new model to speed up theattack detection, to reduce error notification rate and finally to enhance system's efficiency. In this study, for enhancing efficiency of theneural network in detecting intrusions, a genetic algorithm was used for selecting features. The suggested model was examined and assessedon NSL-KDD dataset which is the modified version of the KDD-CUP99. The experimental results indicate that the suggested model is veryefficient in enhancing precision and recall of attack detection and reducing the error notification rate and also is able to offer more accuratedetections in contrast to the basic models
