As the world becomes more digitalized, the potential for attacks increases, therefore, effective techniques for intrusion detection on network are needed. In this study, the authors propose a two steps approach. First, the Correlation-based Features Selection as a feature evaluator based on Particle Swarm Optimization is used to select the relevant features. This evaluator is compared with other evaluators. Second, the Random Forest algorithm is used to classify attacks in a network. A comparative study is also performed conducted with different classifiers such as Na¨ıve Bayes, Stochastic Gradient Descent, Deep Learning, k-Nearest Neighbors and Support Vector Machine. Experiments were conducted on the NSL-KDD database and the results show an efficiency of 98.78% for binary classification. The performance results obtained show that the proposed technique performs better than other competing techniques.

Classification; Feature selection; Intrusion detection system; Machine learning; NSL-KDD data set; Particle swarm optimization; Random forest