摘要
免疫聚类算法中网络刺激与抑制阈值参数决定了聚类精度和网络的可伸缩性,现有的免疫聚类算法中这些阈值选择采取定值策略,根据问题的特性和操作者的经验确定,算法的泛化能力较差。提出了一种自适应免疫聚类算法,阈值从动态变化的网络结构特征中获取,在网络进化过程中,阈值始终跟随网络内在结构变化而自适应调整,因而获得的最终网络结构更符合原始数据的内在特性,并很好地解决了算法对问题的依赖性问题,提高了算法的泛化能力。仿真实验表明了该算法的有效性。
In the artificial immune algorithm for clustering analysis, its suppression and stimulate thresholds determine cluster precision and network population scale. Theses thresholds adopt fixed value that is decided according to the problem characteristic itself and user＇s experience. However, this modus operandi results in narrow application situation and is dependent heavily on problem characteristic itself. Therefore, an adaptive artificial immune algorithm for clustering was proposed. This algorithm could achieve final network structure well matching the crude data feature and relieve the dependence on problem characteristic itself, because its thresholds were obtained from the dynamic immune network structure and were adapted ＂well to the entire network structure during the process of evolution. Experimental results demonstrate its effectiveness.
出处
《计算机应用》
CSCD
北大核心
2006年第8期1943-1946,共4页
journal of Computer Applications
基金
广东省十五规划资助项目(05SJY009)
广东省科技厅科技攻关项目(2005B10201006)
中山大学青年教师科研基金资助项目(2005350001131100)
关键词
数据挖掘
人工免疫系统
聚类分析
自适应
data mining
artificial immune system
clustering analysis
adaptability
