摘要
针对基于单层SOM神经网络的入侵检测系统计算量大、误报率高的问题,利用SOM网络中相似模式激活神经元的物理位置邻近的特点,根据输入模式的类型,对激活的神经元进行划分,并把记录的基本特征和推导特征结合起来,对记录进行分类.研究结果表明,较小的特征子集能使系统更快地对数据进行分类,与传统的利用单层SOM神经网络方法相比,该方法计算量小、误报率低.
The characteristic that similar patterns activate the neighboring neurons is used to classify the records. The activated neurons were classified according to the types of the records. Using appropriate combination of basic features and derived features is proposed to reduce false positive rate and computation of single-layer SOM-based intrusion detection system. The result shows that the smaller subset of features can make the system classify the records faster. This method has the advantage of less computation and lower false positive rate comparing that of using traditional single layer SOM neural network.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2007年第7期5-7,共3页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
关键词
入侵检测
神经网络
特征
自组织特征映射
intrusion detection
neural network
feature
self-organizational characteristic map （SPM）
