改进模糊聚类算法及其在入侵检测中的应用

针对将数据集隶属度概率和为1的条件用于模糊性事件时,影响聚类的正确率的情况,在不确定理论的基础上,研究了数据隶属度问题,提出一种新的基于隶属关系不确定的可能性改进模糊聚类算法.该算法在迭代过程中将聚类的可能隶属度与不确定性隶属度引入到目标函数中,使得样本中的元素不局限于仅属于一个聚类,与现有的聚类算法相比具有更好的聚类结果.通过在KDDCUP99数据集上实验,验证了该算法在入侵检测中的检测率为95.8%, 分别高于K-均值算法的检测率(60.4%)和FCM算法的检测率(64.6%).

Improved fuzzy clustering algorithm and its application to intrusion detection

Since the condition that the sum of possible membership degree of data set is 1 will make negative effect on the correction ratio of fuzzy clustering in fuzzy events,some research on the membership degree of data is conducted and a new improved fuzzy clustering algorithm using the possibility of uncertainty membership degree is presented on the basis of uncertainty theory.Possible membership degree and uncertainty membership degree are introduced into object functions in iterative processing of this algorithm to make the element sample not longer belong to one cluster only,which leads to more preferable results than current clustering algorithms.Experiments on dataset KDDCUP99 testify that the detection rate of this algorithm is up to 95.8% in intrusion detection,which is better than the rate of K-means(60.4%) and fuzzy C-means(FCM)(64.6%).