基于最大内聚度基准的加权投票聚类集成

Weighted voting clustering ensemble based on maximum cohesion

提出一种基于投票的聚类集成方法.通过分析聚类结构与聚类准确率的关系,将内聚度最高的聚类成员作为重新标记的基准以实现簇标记的统一;同时,根据数据点在不同聚类成员中与所划分簇中心的距离确定权值,最终实现加权投票.实验结果表明,该算法在准确率和稳定性上均有较大提高.

A voting-based clustering ensemble method is presented. By analyzing the relationship between the clustering structure and accuracy, the highest cohesive cluster member is considered as the benchmark of relabel algorithm to unify cluster labels. Then the voting weights are determined by the distance from cluster centers which data points in different cluster members are divide into. The experimental results show that, the proposed algorithm is greatly improved in accuracy and stability.