自动属性加权的K-调和均值聚类算法

发布时间：2018-05-02 13:28

本文选题：K-调和均值 + 聚类　；参考：《计算机应用与软件》2016年11期

【摘要】：针对K-调和均值算法中距离度量将所有属性视为相等重要而存在的不足,提出一种利用自动属性加权的改进聚类算法。在算法的目标函数中,用加权欧氏距离替代传统的欧氏距离,并证明了使得算法能够收敛的属性权重更新机制。为进一步提高聚类性能,将粒子群算法融入到改进的属性加权聚类算法中以抑制其陷于局部最优,其中采用聚类中心和属性权重的值同时表示粒子的位置进行寻优。在UCI数据集的测试结果表明,该算法的聚类指标平均提高了约9个百分点,具有更高的聚类准确性和稳定性。
[Abstract]:An improved clustering algorithm based on automatic attribute weighting is proposed to overcome the disadvantage that the distance measure in the K- harmonic mean algorithm considers all attributes to be equal. In the objective function of the algorithm, the traditional Euclidean distance is replaced by the weighted Euclidean distance, and the attribute weight updating mechanism that makes the algorithm convergent is proved. In order to further improve the clustering performance, the particle swarm optimization (PSO) algorithm is incorporated into the improved attribute weighted clustering algorithm to restrain it from falling into the local optimum, in which the values of the clustering center and the attribute weight are used to represent the location of the particles for optimization at the same time. The test results in UCI data set show that the clustering index of the algorithm increases by about 9 percentage points on average and has higher clustering accuracy and stability.
【作者单位】：江南大学物联网工程学院;
【基金】：江苏省自然科学基金项目(BK20140165)
【分类号】：TP311.13

【相似文献】