An Intrusion Detection Model Based on Random Tree Algorithm with Dimensionality Reduction and Oversampling

Abstract

With the advancement of the university information process, more and more application systems are running on the campus network, and the information system becomes larger and more complex. With the rapid growth of network users and the popularization and deepening of computer knowledge, the campus network has been transformed from an experimental network for education and scientific research into an operational network that attaches equal importance to education, scientific research, and service. As the most important transmission carrier of digital information, how to ensure its security has become an urgent issue for colleges and universities. Therefore, this paper uses advanced intrusion detection technology to design the corresponding model to solve the security problem of the campus network. When traditional machine learning algorithms are trained on network intrusion data sets, they are prone to problems such as too many feature dimensions, overfitting, and imbalance of data sets, which lead to lower accuracy and low time efficiency of intrusion detection algorithms. In order to solve the above problems, this paper proposes an intrusion detection model based on Extra Trees, which uses linear discriminant analysis to reduce the dimension of data, then uses oversampling to reduce the influence of imbalance of sample categories of the network intrusion dataset, and finally uses Extra Trees algorithm to train the model. The experimental results show that after LDA reduction and oversampling, using the Extra Trees classification model can improve the overall recognition performance of imbalanced data sets under multi-classification and satisfy the network intrusion detection.