摘要
数据集的聚类边界不清晰时,人工免疫网络聚类使用最小生成树确定聚类数的依据往往不足。分析了问题存在的原因,提出一种基于人工免疫网络的半监督聚类算法。该算法一方面在抗体克隆操作中嵌入数据集的先验信息,抑制位于聚类边界区域抗体的激活能力,从而保证记忆网络能更清晰的反映数据集中各聚类原型的结构;另一方面,将先验信息用于后期记忆网络的最小生成树分割,有效缓解了因聚类边界模糊而无法获得正确的聚类结果。仿真结果表明,该算法对聚类边界不清晰的数据集可获得较精确的聚类结果,同时运行效率也明显改善。
When the data do not exhibit clear clustering structure, artificial immune network clustering algorithm has not usually enough evidence to determine the number of clusters using the minimum spanning tree criteria. Here we address this problem and propose a novel semisupervised clustering algorithm for artificial immune network. The proposed algorithm has two major contributions. Firstly, by embedding prior information of the data set into the cloning of antibodies, it can effectively constrain the activity of antibodies located at the cluster boundaries so as to ensure that the memory network can reflect the structures of the clusters more clearly. Secondly, by utilizing prior information in minimum spanning tree based segmentation in the memory network, we can tackle difficult cases with vague cluster boundaries. Simulation results show that the proposed algorithm can obtain accurate clustering results for difficult data sets with improved efficiency.
出处
《计算机系统应用》
2011年第12期99-104,共6页
Computer Systems & Applications
关键词
半监督聚类
人工免疫网络
克隆选择
成对约束
semi-supervised clustering
artificial immune network
clonal selection
pairwise constraints