EARLY DETECTION OF NETWORK INTRUSIONS USING MACHINE LEARNING TECHNIQUES: A SYSTEMATIC LITERATURE REVIEW
Abstract
The rising complexity and sophistication of threats in network traffic pose significant challenges that require precise and reliable Intrusion Detection Systems (IDSs). In recent years, these systems have gained considerable attention due to their continuously improving performance. The early detection of breaches is critical to limit potential damage, and Machine Learning (ML) methods have shown efficacy in improving IDSs. This review considers 30 recent research papers identified through a systematic screening process. This review seeks to thoroughly analyze and synthesize contemporary research on the classification of supervised and unsupervised ML approaches for the early detection of network breaches. The primary purpose of this review is to provide a thorough, multi-dimensional analysis of approaches for the early detection of intrusions in IDSs. The study emphasizes that early detection relies not only on high accuracy but also on computational efficiency and low latency. Accordingly, it analyzes the influence of feature selection on model effectiveness and speed, methods for handling data imbalance, and commonly used datasets and evaluation metrics. Our findings reveal that Random Forest (RF) and Extreme Gradient Boosting (XGBoost), in particular, demonstrate the optimal balance between high accuracy and low latency, making them superior for early detection compared to Deep Learning (DL) models which often incur higher computational costs. Our study helps researchers identify the most appropriate techniques for early detection of breaches using ML techniques, in addition to providing a research roadmap for future studies in this field