基于深度学习的入侵检测数据分类研究

Research on Intrusion Detection Data Classification Based on Deep Learning

摘要: 针对由于入侵检测数据集中数据类别不平衡，而导致的检测分类准确率低的问题，设计一种基于生成对抗网络(GAN)和深度森林结合的入侵检测模型。首先，基于生成对抗网络独有的对抗思想，通过原数据类的分类结果，筛选出需要生成的类别，生成数据集中缺少的数据，缓解数据集不平衡的问题；然后，针对网络流量特征复杂与深度森林模型数据处理成本高的矛盾，设计了基于主成分分析和线性判别算法结合的特征提取方法，解决了深度森林模型中的数据计算冗余问题，提高了数据传递与处理能力。实验结果证明，所提方法的分类检测精度达到了96%，其中少数类数据的检测精度比没有平衡前提高了30%。

Abstract: Aiming at the problem of low detection and classification accuracy due to the imbalance of data categories in the intrusion detection dataset, an intrusion detection model based on the combina-tion of generative adversarial networks (GAN) and deep forest is designed. .First of all, based on the adversarial characteristics of generated adversarial networks, the classes that need to be generated are screened out through the classification results of the original data and the missing data in the dataset is generated to alleviate the problem of dataset imbalance. Then, aiming at the contradic-tion between the complex network traffic characteristics and the high data processing cost of the deep forest model, a feature extraction method based on the combination of principal component analysis and linear discriminant analysis is designed. It solves the data calculation redundancy problem in the deep forest model and improves the data transmission and processing capabilities. The experimental results show that the classification detection accuracy of the proposed method reaches 96%, and the detection accuracy of the minority class data is 30% higher than that without balance.