Application of Machine Learning for Anomaly Detection in Network Traffic to Enhance Cybersecurity

Abstract

The increasing volume and complexity of network traffic have created significant challenges for cybersecurity, particularly in identifying abnormal activities that may indicate security threats. Machine learning has emerged as an important approach for network anomaly detection because it can identify complex traffic patterns and support automated threat detection. This study aims to examine the application of machine learning for anomaly detection in network traffic and analyze its contribution to cybersecurity. The study employed a structured literature review and comparative qualitative synthesis based on 27 scientific publications published between 2024 and 2026. The analysis focused on conventional machine learning, deep learning, ensemble learning, hybrid approaches, real-time detection, adaptive detection, and scalable network anomaly detection. The empirical evidence shows that algorithm performance varies according to the dataset and evaluation context. Pai et al. (2025) reported accuracies of 98.28% for K-nearest neighbors, 97.00% for random forest, 96.62% for gradient boosting, 87.95% for support vector machine, and 85.59% for logistic regression on the CICIDS2017 dataset. Their hybrid ensemble stacking approach achieved 98.79% accuracy. Ness et al. (2025) reported 1.00 training accuracy and 0.85 test accuracy for LightGBM, demonstrating the importance of evaluating model generalization. The study concludes that machine learning can strengthen network anomaly detection, but model selection must consider data characteristics, feature complexity, class imbalance, scalability, and real-time processing requirements.