摘要
以入侵检测系统中的分类器设计为例,研究分类器训练样本选择问题.提出了一种大规模数据集的训练样本选择方法.首先通过聚类将训练数据划分为不同的子集,缩小搜索范围;然后根据聚类内离散度和样本的覆盖区域选择样本,保留每个聚类的边界样本,删除内部样本.仿真结果证实,由于保留了典型样本,减少了训练样本数量,从而保证了分类器的性能且训练效率较高.
Taking the example of designing classifier in intrusion detection system, the selection of training samples for classifier is studied. A new method is proposed for sample selection in large data set. First, it will reduce the size of selection problem via clustering, select samples according to the with-in cluster scatter value and coverage area of a sample. And it will retain boundary samples and discard most of the interior ones in each cluster. Experiment result shows that as reserving typical samples and reducing training samples, the generalization performance and training efficient of the classifier are guaranteed.
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
2006年第4期77-80,共4页
Journal of Beijing University of Posts and Telecommunications
基金
国家自然科学基金项目(60475007)
关键词
样本选择
离散度
覆盖区域
边界样本
sample selection
scatter
coverage area
boundary samples