CAJ | 학술논문

针对网络入侵检测中的高维数据处理问题，提出基于半监督降维技术和BP神经网络的入侵检测方法，该方法主要有两个优点：实时性更高；训练样本标记工作量更小。对半监督降维技术背后的数学原理进行解释，并论述其在网络入侵检测中应用的适用性。对比实验表明：在少量标记样本和大量未标记样本的支持下，半监督降维技术能够在降低维数的同时保持入侵检测性能，从而大幅降低入侵检测的训练和检测时间。
침대망락입침검측중적고유수거처리문제，제출기우반감독강유기술화BP신경망락적입침검측방법，해방법주요유량개우점：실시성경고；훈련양본표기공작량경소。대반감독강유기술배후적수학원리진행해석，병논술기재망락입침검측중응용적괄용성。대비실험표명：재소량표기양본화대량미표기양본적지지하，반감독강유기술능구재강저유수적동시보지입침검측성능，종이대폭강저입침검측적훈련화검측시간。
To solve the problem of high-dimensional data processing in network intrusion detection,we propose an intrusion detection method which is based on semi-supervised dimensionality reduction and BP neural networks.It has two advantages mainly:higher real-time performance and lower cost in training sample labelling.The mathematical principle of semi-supervised dimensionality reduction is analysed in the paper.We also discuss its adaptability in network intrusion detection.Contrastive experiment demonstrates that with the support of very few labelled samples and abundant unlabeled samples,the semi-supervised dimensionality reduction can maintain the intrusion detection performance while reducing the dimensionality,therefore it greatly reduces the training and detection times of intrusion detection.