两阶段混合粒子群优化聚类

为解决数据集样本维数较高时已有粒子群优化K均值算法计算速度较慢且聚类结果不稳定的问题,利用第1阶段聚类层次凝聚聚类获得准确率较高的子簇集合,作为粒子群优化K均值聚类算法初始聚类中心的搜索空间,进行第2阶段聚类.提出了一种简化的粒子编码方法,以减小样本维数对计算复杂度的影响;引入混沌的思想,以保持粒子种群的多样性,从而避免粒子群优化算法可能出现的早熟现象.通过两阶段聚类,有效地融合了粒子群优化、层次聚类与划分聚类算法的优点.在多个UCI数据集上的聚类结果表明,与几种对比算法聚类结果的最优值相比,其纯度分别提高了1%~8%,且耗时减少50%以上. 

Two-Step Hybrid PSO-Based Clustering Algorithm

In order to solve the problems of the existing PSO (particle swarm optimization) K-means algorithms, i.e., their calculation speeds are slow and the clustering results are unstable when samples have a high dimension, some high-quality sub-clusters were generated by hierarchical agglomerative clustering. These sub-clusters were used as the search space of candidate centroids of the PSO K-means. In order to reduce the computational complexity when the dimension of a sample is high, a simplified particle encoding method was proposed. In addition, chaotic idea was introduced to keep the diversity of particle swarm to avoid premature. By two-step hybrid clustering the advantages of the hierarchical clustering, the partitioning clustering and the PSO were combined. The experimental results on several UCI data sets show that compared with the best results of several contrastive algorithms, the purity of its clustering result increases by 1% to 8% and the consuming time reduces by 50% at least.