基于深度学习和R-Drop正则的入侵检测模型

Intrusion Detection Model Based on Deep Learning and R-Drop Regularization

在入侵检测分类任务中,传统的机器学习模型其性能往往不能达到较好的效果,深度学习技术泛化能力更强,因此研究深度学习算法并应用到入侵检测检测系统中是有十分有意义的。论文经过研究,针对网络流量2分类问题,提出了一种基于FNet分类模型——TFN;针对网络流量多分类问题,提出了一种基于R-Drop正则的深度学习多分类模型。论文使用入侵检测数据集NSL-KDD作为实验数据,实验结果表明,在NSL-KDD的数据集上,提出的2分类模型效果优异,准确度达到99%;而提出的多分类方法,与普通训练方法相比也提升了1%~2%的准确度。

In the task of intrusion detection classification,the performance of traditional machine learning models often cannot achieve good results,and the generalization ability of deep learning technology is stronger. Therefore,it is of great significanceto study deep learning algorithm and apply it to the intrusion detection system. After research,aiming at the problem of network traffic 2 classification,this paper proposes a classification model based on FNet,it is TFN. Aiming at the problem of multi-classification of network traffic,a deep learning multi-classification model based on R-Drop regularization is proposed. This paper uses theintrusion detection data set NSL-KDD as the experimental data. The experimental results show that the proposed 2 classificationmodel has an excellent effect and accuracy of 99.99% on the NSL-KDD data set. The proposed multi-classification method also improves the accuracy by 1%~2% compared with the ordinary training method.