基于强化学习的特征提取方法在攻击识别中的应用

Application of Feature Extraction Method Based on Reinforcement Learning in Attack Recognition

针对工控数据集特征数量较大时的分类准确率较低和训练时间较长等问题,提出了一种采用强化学习来进行特征提取对数据集进行预处理的方法.首先,通过强化学习确定过程矩阵和决策矩阵,再根据决策矩阵进行特征提取,获得预处理数据集.将NSL-KDD、自建数据集和密西西比数据集的原始数据集、PCA和强化学习后的数据集分别用神经网络和SVM训练并进行分类,实验结果表明,该方法适用特征关联度较低的数据集进行神经网络训练.强化学习能有效提高分类的准确率、精确率等指标,并减少运行时间,提高效率.

Aiming at the low classification accuracy and long training time of industrial control datasets,this paper proposes a feature selection method to preprocess the dataset based on reinforcement learning.Firstly,it builds a process matrix and a decision matrix by reinforcement learning.Then it extracts the features by decision matrix and gain processed dataset.These processed datasets,such as the NSL-KDD,own created datasets,the original data of the Mississippi datasets,and the datasets disposed by the PCA and reinforcement learning,which are trained and classified by SVM and neural network.The result shows that this method is appropriate on datasets whose features are not relative.Reinforcement learning can increase indexes such as accuracy and precision of classification,decrease training time,and improve efficiency.