Assessing the Effectiveness of Machine-Learning Approaches for Detecting Network Attacks: An Empirical Evaluation on NSL-KDD

Intrusion detection in computer networks is critical for safeguarding sensitive data and maintaining system integrity. This research evaluates the performance of multiple machine-learning models, including Support Vector Machine (SVM), Naive Bayes (NB), Random Forest (RF), Multilayer Perceptron (MLP), and XGBoost, for detecting network intrusions using the NSL-KDD dataset. The dataset, comprising 148,517 network traffic instances with 41 features, represents a diverse range of attack types, including DoS, Probe, U2R, and R2L. Preprocessing steps such as normalization, encoding, and feature selection ensure model-ready inputs. Experimental results demonstrate that ensemble and deep-learning-based models outperform traditional classifiers, with XGBoost achieving the highest accuracy, precision, recall, and F1-score. The findings highlight the effectiveness of modern machine-learning approaches in capturing complex attack patterns and improving intrusion-detection performance.