Improving Intrusion Detection Using Machine Learning Algorithms with Feature Selection based on Information Gain
Abdellah Mazighi, Lahoucine Ballihi, Ghizlane Orhanou · 2023
The expansion of the Internet and mobile communications has created a very large cyberspace. Thus our cyberspace has become very vulnerable to various cyber-attacks emanating from hackers and cybercriminals using increasingly sophisticated and prolonged attacks. Actually, Feature Selection and classification is applied in various domains like business, medical, cyber-security and media field. Huge datasets with a high number of attributes usually contain several insignificant features. In data analysis, those irrelevant features are time consuming, affect the efficiency and computational complexity and increase the amount of resource usage. They affect badly the machine learning performances. The main part of this article consists of many experiments we have conducted on a 20% of the CICIDS-2017 dataset. We have used four machine learning classifiers (Naive Bayes (NB), AdaBoost (AdB), Random Tree (RT) and Random Forest (RF)) and measured their performances before and after feature selection using Information Gain. We have divided the 77 attributes into four groups according to their weights and performed intrusion detection with different machine learning algorithms on each group of attributes. The results have showed that this method improves some metrics such as the global detection, execution time, the time required to learn the model.