Optimization of Class Imbalance Techniques in Machine Learning Models for Network Intrusion Detection

Machine learning based network intrusion detection has been well developed. The challenge of imbalanced datasets in network intrusion detection has been designed for a while. This paper investigates three class imbalance handling techniques, including random sampling (ROS), synthetic minority sampling technique (SMOTE), and a customized SMOTE approach (Strategy SMOTE) in the UNSW-NB15 dataset. By combining these methods with machine learning algorithms such as Random Forest, XGBoost, LightGBM, and Multi-Layer Perceptron, we demonstrate improvements in model performance, particularly in detecting minority attack classes. Our findings highlight the effectiveness of optimized imbalance handling in improving the reliability of intrusion detection systems.