【摘要】 【目的】针对改进的可能性C均值聚类方法(IPCM)运算效率低,难以处理复杂数据结构的问题,提出了一种约束的改进可能性C均值聚类方法(CIPCM).【方法】CIPCM方法采用多项式核将特征向量映射到一个隐性特征空间,便于处理复杂的数据结构;引入两个成对约束集合,降低聚类迭代次数,提高运算效率和抗干扰能力.实验采用国际公认的UCI公共测试数据集,并用错分率指标评测了目标分类性能.【结果】CIPCM方法的聚类错分率低,对噪声的鲁棒性强.【结论】CIPCM运算效率比高于改进可能性C均值聚类方法.更多还原

【Abstract】 【Objective】To solve the problem that improved possibilistic C-means method has low computational efficiency and is hard to deal with complex data structures.【Method】A constrained improved possibilistic C-means method was proposed.The new method used polynomial kernel to map the feature vector to an implicit space,in order to easily deal with complex data structures and introduced two pairwise constrain set to reduce the number of iteration in the process of data clustering to improve the computational efficiency and anti-interference ability.【Result】The experiment was implemented on a well-known public testing dataset called UCI,and used misclassification rate to evaluate the performance of object classification.The results showed that the new method had low misclassification rate and strong robustness to noise.【Conclusion】The computational efficiency of the new method is higher than that of the improved possibilistic C-means clustering method.更多还原