基于互信息和轮廓系数的聚类结果评估方法  

Protocol Clustering Evaluation Method Based on Mutual Information and Contour Coefficient

作　　者：尹世庄 王韬 谢方方 刘丽君 曲直 张斌 YIN Shizhuang;WANG Tao;XIE Fangfang;LIU Lijun;QU Zhi;ZHANG Bin(Shijiazhuang Campus of Army Engineering University, Shijiazhuang 050003, China;Army Engineering University, Nanjing 210000, China)

机构地区：[1]陆军工程大学石家庄校区,石家庄050003 [2]陆军工程大学,南京210000

出　　处：《兵器装备工程学报》2020年第8期207-213,共7页Journal of Ordnance Equipment Engineering

摘　　要：在采用改进k-means对未知二进制协议聚类的基础上,引入调整互信息和轮廓系数两种参数,分别对聚类的簇内聚类效果和整体聚类效果进行评估,并以真实的二进制协议数据为例,验证了评估的有效性。实验表明,基于调整互信息的评估方法类簇中样本判定的准确率大于90%,并且与数据报文真实的类别相符;基于轮廓系数的评估方法对不同数据集聚类结果的评估与聚类结果准确率的分布也相一致。该方法运算速度快,并且准确率优于代表点法,更符合二进制协议特征。On the basis of clustering unknown binary protocols with improved k-means,two parameters of adjusting mutual information and outline coefficient were introduced to evaluate the intra-cluster clustering effect and the overall clustering effect of clustering.Taking the real binary protocol data as an example,the effectiveness of the evaluation was verified.The experimental results show that the accuracy of sample determination in the cluster based on adjusting mutual information is more than 90%,and it is consistent with the real category of data message.The evaluation of different data clustering results by the evaluation method based on contour coefficient is also consistent with the distribution of clustering accuracy.This method has the advantages of fast operation speed and better accuracy than the representative point method,and is more in line with the characteristics of binary protocol.

关 键 词：二进制协议 聚类 互信息 轮廓系数 评估 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]