Intrusion detection system (I.D.S) is an essential component, which enhances the security of computer systems by actively detecting all forms of attack at the early stages. The main use of IDS is the monitoring of the network traffics and analyzing the behavior of the users in searching for any abnormal activity or attack signature for real-time intrusion detection. The main weakness in any IDS is their inability to offer adequate sensitivity and accuracy; coupled with their inability to process enormous data. To address these issues (such as the increasing traffic, huge behavior profiles, large signature databases, and the inability of differentiating normal behaviors from the suspicious ones), several algorithms have been developed. Hence, the main aim of this work is to choose the differentiating features for the development of an optimal machine learning algorithm which can offer high detection rates, fast training, and testing processes offline. The proposed machine learning model contains a feature selection algorithm (wrapper type) which is based on the integration of the Binary Firefly algorithm enhanced for feature selection by crossover operator taking from the genetic algorithm, called (GA-FA) with the Naïve Bayesian Classifier (NBC). The performance of the proposed model was tested on NSL_KDD data sets prepared by the MIT Lincoln Laboratory. The model testing was based on several experiments and different scenarios (the effect of swarm size, number of iterations, and the Swap). For evaluating the ability to select the minimum number of features with the higher value of classification accuracy, the algorithm worked perfectly and selected a comparable number of features. The model achieved the best average accuracy of 97.011%. In conclusion, the proposed feature selection algorithm has the ability to select the most relevant features which enhance the classification accuracy of the network intrusion detection system.