An Intrusion Detection Model Based on SMOTE and Convolutional Neural Network Ensemble

Abstract

Massive, multi-dimensional and imbalanced network traffic data has brought new challenges to traditional intrusion detection systems (IDSs). The detection performance of traditional algorithms is closely related to feature extractions, which are not effective in the massive and imbalanced data environments. In this paper, we propose an intrusion detection model based on synthetic minority oversampling technology (SMOTE) and convolutional neural network (CNN) ensemble. It converts original traffic vectors into images, designs a CNN structure, and combines SMOTE and CNN ensemble to solve the problem of imbalanced datasets. Using the standard KDD CUP 99 dataset to evaluate the performance of the proposed model and analysing the contribution of features to model decision-making show that the model’s F1 score are better than traditional algorithms in the classes with few samples and the model improves the efficiency of network intrusion detection.