摘要
针对海量的等保测评数据,如何从这些数据中选取适量的数据进行入侵行为分析,提出了根据预测变量对预测目标变量的重要性的特征提取方法。该方法采用importance指数来对预测变量进行等级划分。并选取了一些预处理后的数据运用了两种BP算法--标准BP算法和学习速率自适应调整算法进行了系统仿真预测。通过KDDCup99数据集测试表明,后者相对于前者,其学习训练次数大大降低,学习能力和预测准确率明显提高。
For the mass of security evaluation data, how to analyse a intrusion detection from these data, a method which is ac-cording to the characteristics of importance of predictor variables on forecasts of goal variables extraction has been put forward. The method uses a importance index to carry on the classification of variables. And it selects some data after preprocessing to be used in the system simulation and prediction with two kinds of BP algorithm -the standard BP algorithm and the learning rate adaptive adjustment algorithm. Test by KDDCup99 dataset, the latter is more significantly reduced in the frequency of learning training, and is more clearly increased in learning ability and prediction accuracy rate than the former.
作者
卿江萍
刘志杰
徐洋
QING Jiangping1,LIU Zhi-jie1, XU Yang1 (1.Key Laboratory of Information and Computing in Guizhou Province, Guiyang 550001,China; 2.Network Center of Guizhou Normal University, Guiyang 550001,China)
出处
《电脑知识与技术》
2013年第10期6365-6368,共4页
Computer Knowledge and Technology
关键词
特征提取
标准BP算法
学习速率自适应调整算法
feature extraction
the standard BP algorithm
the learning rate adaptive adjustment algorithm