近邻成分分析下的DDoS攻击检测

DDoS Attack Detection Based on NeighborhoodComponents Analysis

为了有效地识别和缓解网络中的攻击流量,提出了基于SDN的DDoS攻击检测防御框架。针对机器学习特征选择精准度较低和分类准确率不高的问题,设计了基于随机梯度下降的近邻成分分析算法,利用机器学习技术,通过在SDN控制平面中部署决策树算法,实现了攻击流量的判别。实验结果表明:分类准确率达到99.63%,特征选择精度达到97%,验证了所提框架的安全性。

In order to effectively identify and mitigate the attack traffic in the network,an SDN-based framework for distributed denial-of-service(DDoS)attack detection and defense was proposed.In addi‐tion,a stochastic gradient descent-based neighborhood component analysis algorithm was designed to address the problems of low accuracy of feature selection and insufficient classification accuracy in ma‐chine learning.By using machine learning techniques,the decision tree algorithm was deployed in the SDN control plane to discriminate the attack traffic.The experimental results show that the classifica‐tion accuracy reaches 99.63%,and the accuracy of feature selection is 97%,which verifies the security of the proposed framework.