聚类集成时机的确定

Occasion Determination of Clustering Ensemble

使用集成学习技术可以提高聚类性能。在实验中发现,当各聚类成员聚类迭代到中后期时进行集成所得的结果会优于其迭代完全停止时进行集成所得的结果。利用集成网络泛化能力的偏差-方差分解理论对聚类集成过程中的上述现象进行解释,将提高集成网络间泛化能力的早期停止准则应用于聚类集成过程,并提出聚类集成时机的概念。对比实验表明,基于早期停止准则的聚类集成得到的结果较好,且更节约聚类集成的时间,为寻求聚类集成的最佳时机提供了可行性建议和方法。

Ensemble learning technique may improve the clustering performance.In the experiment,we discovered that combining the mid-to-late solutions of cluster members in different initial conditions probably get the better ensemble results than combining the end ones.We used the bias/variance trade-off of generalization ability in ensemble network to explain this phenomenon,applied the early stopping rules to the clustering ensemble and proposed the concept of clustering ensemble occasion.The experimental results show that the performance of clustering ensemble based on the early stopping rules is superior to that based on the end solutions of cluster members,while the former takes less time,thus giving some useful suggestions for seeking the best clustering ensemble occasion.