Weighted Feature Selection for Machine Learning Based Accurate Intrusion Detection in Communication Networks

Network intrusion detection systems work on huge data sets, with large feature sets dominated by noisy data and irrelevant features, resulting in steep degradation in detection accuracy and a steep proliferation in model training and computation time. This work presents a novel method to optimize the feature selection process in machine learning algorithms for accurate detection of intrusion attacks in communication networks. The proposed method targets features with a high impact on the target variable to optimize feature selection and reduction. The CICIDS-2017 data set is used to test the performance of the proposed approach. Results prove the dexterity of the proposed method as it is able to achieve an almost 51% reduction in irrelevant features and increases the detection accuracy of the tuned random forest classifier to 99.9% with an almost 50% reduced model computation time.