海量数据环境下用于入侵检测的深度学习方法

Deep learning method for intrusion detection in massive data

针对传统浅层机器学习方法无法有效解决海量入侵数据的分类问题,提出了一种基于深度信念网络的多类支持向量机入侵检测(DBN-MSVM)方法。该方法利用深度信念网络对大量高维、非线性的无标签原始数据进行特征降维,从而获得原始数据的最优低维表示;利用二叉树构造多类支持向量机分类器,并对获得的最优低维表示进行网络攻击行为识别。最后在KDD’99数据集上进行实验仿真,DBN-MSVM方法可缩短支持向量机分类器的训练时间和测试时间,提高了海量入侵数据的分类准确率。

In order to solve the problem that intrusion massive data is not effectively classified using traditional machine lear-ning methods,this paper proposed an intrusion detection method of multi-class support vector machine based on deep belief nets(DBN-MSVM).Firstly,it employed deep belief nets to reduce the feature dimension of large amounts of nonlinear high-dimensional unlabeled input data,and obtained the optimal low-dimensional representation of raw data.Secondly,it used a binary tree structure multi-class support vector machine classifier to recognize intrusion from the optimal low-dimensional data.Finally,experimental results demonstrate that the DBN-MSVM method can reduce the training time and testing time of support vector machine classifier and raise classification accuracy of intrusion massive data on KDD’99 dataset.