Network Intrusion Detection Model based on Chi-square Test and Stacking Approach
Zheng Xiao, Yu Wang, Luliang Jia, Dapeng Xiong, Jie Qiang · 2020
In recent years, with the development of the Internet, the amount of network data continues to grow. IDS based on traditional machine learning algorithms faces the following problems. On the one hand, massive data may reduce the detection efficiency. On the other hand, IDS based on the single machine learning model shows a trend of diminishing marginal utility. To deal with these aforementioned issues and ensure the security of network, this paper proposes a network intrusion detection model based on chi-square test and stacking approach. Firstly, in order to reduce the difficulty of learning and detection tasks, chi-square test is used to select the effective features to reduce the dimensionality of the dataset. Secondly, to optimize detection accuracy, we build a more powerful intrusion detection model by ensemble learning stacking approach, which integrates multiple machine learning algorithms such as SVM, BPNN, K-Means and XGBoost. At the same time, the K-Fold cross-validation method is used to train this model to reduce possible overfitting. This model is verified using the NSL-KDD dataset. The results obtained from the experiments show that the proposed model performs better in the field of intrusion detection than the single machine learning model. The accuracy, precision and recall reached 97.9%, 95.2% and 98.7%, respectively.