摘要
网络入侵检测系统IDS中,异常数据所占的比例非常小,属于小类样本,却是检测的目标。在AdaBoost算法基础上进行改进,通过对大类样本权重设置阈值,对权值超过阈值的样本进行相应处理,来削弱分类器对大类样本错分的重视程度,减轻下一级训练的负担,从而有效地强化对小类错分样本的学习,提高入侵检测的精度,降低误报率和漏报率。方法在KDD-99数据集上进行实验,并与SVM方法检测结果进行比较,取得了很好的效果。
In network intrusion detection system,abnormal samples are the goal of detection,although the proportion of them is very small.Based on AdaBoost algorithm,this paper sets threshold value to the samples of majority class and processes those whose weight is above the threshold.The improved algorithm can effectively reduce the burden of classifier on next level of training and strengthen the learning to the samples of minority class.The results experimented on KDD-99 sets and comparison with SVM method show the algorithm effectively reduces the false alarm rate and omission rate.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第21期94-96,共3页
Computer Engineering and Applications
基金
安徽省优秀青年人才基金(No.2010SQRL134)
