Machine Learning-Based Intelligent Network Intrusion Detection System Using Random Forest Classification

The fast-paced development of communication infrastructures and integrated cloud systems poses an increased risk of being exposed to any form of cyber threats. The traditional IDS that relies on signatures has certain limitations when it comes to detecting new or zero-day exploits. In order to overcome this crucial limitation, this work proposes a machine learning network security solution that involves the use of the Random Forest classifier. For testing purposes, the NSL-KDD data set is used to detect several types of attacks, namely, DoS, U2R, R2L, and probes. The entire methodological framework consists of various stages: data pre-processing, labeling, scaling, feature selection, and testing against other supervised machine learning approaches such as Naïve Bayes, Decision Tree, K-Nearest Neighbors, and Logistic Regression. It has been proven empirically that the proposed Random Forest model outperforms other algorithms, with its superior accuracy equalling 98.3%. Finally, the presented approach allows for developing a system that produces the fewest possible false positives, offers scalability, and can accurately track multiple classes of attacks.