Effective Intrusion Detection in Imbalanced Network Traffic using XGBoost and LightGBM

With the continuous evolution of network attack methods, traditional Intrusion Detection Systems (IDS) face increasing challenges. To improve the accuracy and responsiveness of IDS, this paper proposes a network traffic classification approach based on two machine learning algorithms: XGBoost and LightGBM. We train and evaluate both models using several publicly available network traffic datasets to compare their performance in handling different types of network attacks. The models are evaluated using metrics such as accuracy, precision, recall, and F1 score, while confusion matrix analysis assesses performance across various attack categories. Experimental results show that XGBoost outperforms LightGBM in most attack categories, especially in detecting DDoS attacks, where it demonstrates higher precision and recall. On the other hand, LightGBM offers advantages in training speed and memory consumption, making it more suitable for large-scale datasets. Based on accuracy, efficiency, and resource consumption, this paper provides two effective machine learning methods for network intrusion detection and discusses their application scenarios and potential improvements.