CAJ | 학술논문

针对k-medoids算法对初始聚类中心敏感，聚类精度较低及收敛速度缓慢的缺点，提出一种基于密度初始化、密度迭代的搜索策略和准则函数优化的方法。该算法初始化是在高密度区域内选择k 个相对距离较远的样本作为聚类初始中心，有效定位聚类的最终中心点；在k个与初始中心点密度相近的区域内进行中心点替换，以减少候选点的搜索范围；采用类间距和类内距加权的均衡化准则函数，提高聚类精度。实验结果表明，相对于传统的k-mediods 算法及某些改进算法，该算法可以提高聚类质量，有效缩短聚类时间。
침대k-medoids산법대초시취류중심민감，취류정도교저급수렴속도완만적결점，제출일충기우밀도초시화、밀도질대적수색책략화준칙함수우화적방법。해산법초시화시재고밀도구역내선택k 개상대거리교원적양본작위취류초시중심，유효정위취류적최종중심점；재k개여초시중심점밀도상근적구역내진행중심점체환，이감소후선점적수색범위；채용류간거화류내거가권적균형화준칙함수，제고취류정도。실험결과표명，상대우전통적k-mediods 산법급모사개진산법，해산법가이제고취류질량，유효축단취류시간。
For the disadvantages that sensitivity to centers initialization, lower clustering accuracy and slow convergent speed of k-medoids algorithm, a novel k-medoids algorithm based on density initialization, density of iterative search strategy and optimi-zation criterion function is proposed. The Initialization of the algorithm is that, it chooses k cluster centers in the high-density area which are far apart, effectively positioning of the final cluster center. To replace the centers are in the ranges which are proximity to the k-initial centers, to reduce the scope of the search candidate point. Criterion function of equalization based on class density and within-class density weighted is adopted to improve the clustering precision. Experimental results show that this algorithm can improve the clustering quality, shorten the clustering time compared with traditional k-medoids algorithms or other improved algorithms.