基于对抗机器学习的网络入侵特征选择研究

Research on network intrusion feature selection based on anti⁃machine learning

网络入侵特征选择过程中无法有效缩减特征规模,导致特征选择效果不理想,据此该文设计基于对抗机器学习的网络入侵特征选择方法。利用对抗机器学习以及Word2vec改变入侵特征规模,降低冗余特征量。选择对抗入侵节点位置,通过迭代更新获取节点可信度和可用度值,形成入侵特征选择点集,从入侵特征分组中随机选择特征子集,获取网络入侵特征选择结果。由实验结果可知,该方法特征选择结果与理想结果一致,F1分数最高为95分,特征选择效果较好。

In the process of network intrusion feature selection,the feature size cannot be effectively reduced,resulting in the unsatisfactory effect of feature selection.Therefore,a network intrusion feature selection method based on anti-machine learning is designed.The scale of intrusion features is changed by using anti-machine learning and Word2vec to reduce the amount of redundant features.Select the location of anti intrusion nodes,obtain node credibility and availability values through iterative updates,form an intrusion feature selection point set,randomly select feature subsets from intrusion feature groups,and obtain network intrusion feature selection results.The experimental results show that the feature selection result of this method is consistent with the ideal result,with the highest F1 score of 95,and the feature selection effect is good.