Comparative Analysis of Multiclass Classification Machine Learning Models for Cybersecurity Intrusion Detection

Recently, the use of the Internet and computer networks in general, has increased exponentially, leading to a high demand for cybersecurity to protect against all kinds of network attacks that are constantly evolving. Nowadays, Machine Learning (ML) is implemented in various cybersecurity tools, Intrusion Detection System-based ML brings more capabilities to improve the detection of cyber attacks. This study aims to perform a comparative analysis of a multiclass classification problem for cybersecurity attack detection. The comparative analysis is performed using six different machine learning algorithms: Naïve Bayes, Decision Tree, Random Forest, Support Vector Machine, eXtreme Gradient Boosting, and Multi-Layer Perceptron, which are applied to the full NSL-KDD dataset, and three other subsets of datasets to confirm and verify the findings in terms of precision, accuracy, training time, and testing time. In all dataset subsets we worked on in addition to the NSL-KDD dataset, eXtreme Gradient Boosting significantly beat the other algorithms. From all the experimental results, it is concluded that XGBoost is a plausible choice for an intrusion detection system in terms of all the metrics compared to the other ML algorithms discussed.