摘要: 传统集中式异常检测方法需要耗费大量的网络资源和计算时间。为此,提出一种基于模型共享的分布式异常检测方法。利用多数投票、边界扩展、平均叠加以及距离加权这4种集成学习方法得到全部局部模型,通过交换本地数据挖掘模型的方式实现数据共享,构造总体的集成式学习模型。实验结果表明,该模型能从全局的观点检测异常,减少集中式检测所需的数据传输量,有效地保护数据的隐私性。
关键词:
数据挖掘,
集成学习,
异常检测,
数据共享,
局部模型,
检测性能
Abstract: Traditional centralized anomaly detection methods take a lot of network resources and computing time. Therefore, a distributed anomaly detection method based on model share is proposed. The method firstly employs majority vote, boundary extension, average stack and weighted distance to obtain all local models, and then exchanges local data mining models to realize data sharing, so that the overall integrated learning model is established. Experimental results show that the proposed method can detect anomaly from the global view, reduce test data transmission of overall integrated learning model and effectively protect data privacy.
Key words:
data mining,
ensemble learning,
anomaly detection,
data sharing,
partial model,
detection performance
中图分类号:
陈小玉, 李晓静, 周绪川. 基于模型共享的分布式异常检测方法[J]. 计算机工程, 2012, 38(11): 262-263,267.
CHEN Xiao-Yu, LI Xiao-Jing, ZHOU Xu-Chuan. Distributed Anomaly Detection Method Based on Model Share[J]. Computer Engineering, 2012, 38(11): 262-263,267.