Machine Learning Techniques for Network Intrusion Detection: A Comparative Review of Algorithms, Datasets, and Performance Metrics (2020–2025)

Abstract

This paper presents a comparative review of machine learning techniques used in network intrusion detection systems by analyzing fourteen studies published between 2020 and 2025. The review was conducted according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses )PRISMA) guidelines, and the studies were identified using Google Scholar with a search string combining terms related to machine learning, intrusion detection, and performance evaluation. The studies were selected based on their application of at least one machine learning algorithm, the presence of quantitative performance metrics, and publication in English within the specified period. The reviewed studies employed algorithms that fall into three main categories: traditional models such as Random Forest, Decision Tree, and Support Vector Machine; deep learning models such as Long Short-Term Memory and Deep Belief Networks; and ensemble methods such as XGBoost and AdaBoost. Of the thirteen studies that reported a single summary accuracy value, the average accuracy was 96.42%, ranging from 83.27% to 99.99%. The Random Forest algorithm was the most frequently tested, appearing in ten out of fourteen studies. The NSL-KDD dataset was the most frequently used, appearing in six studies. However, nine out of fourteen studies did not report an F1-Score, thus limiting the possibility of direct comparison between the results. The results indicate that traditional models and ensemble methods remain highly competitive compared to deep learning methods when applied to standard datasets. Furthermore, the choice of dataset significantly impacts the reported performance levels. The study recommends that future research focus on using newer datasets that are more representative of real-world environments, adhering to consistent reporting of multiple performance metrics, and developing standardized evaluation frameworks that allow for more reliable comparisons between different studies.