摘要
文章针对增量网络入侵数据属性冗余导致的传统学习算法效率低、检测精度差等问题,提出一种基于粗糙集属性约简的在线序贯极限学习机(AR-OSELM)方法。该方法首先对入侵数据采用粗糙集正域和分辨矩阵的方法获得属性核,筛选出无冗余属性的特征集合,然后使用在线序贯极限学习机作为分类算法进行分类。仿真实验结果表明,与BP、ELM及HELM神经网络算法相比,AR-OSELM算法对增量数据的学习和训练效率更高,入侵检测准确,误报率较低。算法有较好的泛化能力,为网络入侵检测提供了一种新的方法。
Considering the low learning efficiency and poor detection precision of the traditional learning algorithms caused by the redundant attributes of the incremental network intrusion data, this paper proposes an online sequential extreme learning machine algorithm based on attributes reduction in rough set（AR-OSELM）. Firstly, the attribute kernels are obtained by using the methods of rough set positive domain and discernibility matrix on intrusion data,thus characteristic collections of non-redundant attributes are obtained. Then using the online sequential extreme learning machine as the classification algorithm to classify the data sets. The results of the simulation experiment show that that the AR-OSELM algorithm is more efficient in learning and training incremental data and has lower error rates with comparison to BP, ELM and HELM algorithms. The AR-OSELM algorithm has better ability of generalization than other tradition algorithms which provides a new method for network intrusion detection.
作者
魏书宁
陈幸如
焦永
王进
WEI Shuning;CHEN Xingru;JIAO Yong;WANG Jin(College oflnformation Science andEngineering,Hunan Normal University,Changsha Hunan 410006,China;Internet of Things Technology andApplication Key Lab,Hunan Normal University,Changsha Hunan 410006,China)
出处
《信息网络安全》
CSCD
北大核心
2018年第6期1-6,共6页
Netinfo Security
基金
国家自然科学基金[61472437]
湖南省教育厅一般项目[531120]
湖南师大自然科学研究项目[物160432]
