Intrusion Detection Model Based on Autoencoder and XGBoost
Yunxiang Kang, Minsheng Tan, Lin Ding, Zhiguo Zhao · Journal of Physics Conference Series · 2022
Abstract In recent years, machine learning algorithms have been extensive used for intrusion detection field. At the same time, these algorithms still suffered from low accuracy due to data imbalance. To improve accuracy of detection, an intrusion detection model based on Autoencoder (AE) and XGBoost (IDAE-XG) is proposed. The training algorithm and detection algorithm related to IDAE-XG are given. IDAE-XG constructs the training set with preprocessed normal data. Data preprocessing includes feature selection and feature grouping. Through detection, XGBoost is used to predict results, which effectively improves prediction accuracy. The superiority of the proposed IDAE-XG is empirically demonstrated with extensive experiments conducted upon CSE-CIC-IDS2018. The experimental comparison show that IDAE-XG performs better than the KitNet model in the test, and has achieved a great improvement in accuracy and recall rate.