基于粒计算的粗糙集聚类算法

针对传统K-means聚类算法初始聚类中心随机选取、不能处理边界对象、效率低、聚类精度低等问题, 提出了一种新的K-means聚类算法。算法引入粒计算理论, 并依据密度和最大最小距离法选择初始聚类中心, 避免初始聚类中心在同一个类中, 结合粗糙集, 通过动态调整上近似集和边界集的权重因子, 以解决边界数据的聚类问题; 最后采用类间距和类内距均衡化准则函数作为算法终止判断条件, 来得到更好的聚类效果。实验结果表明：该算法具有较高的准确率, 迭代次数较少, 并降低了对噪声的敏感程度。

Rough clustering algorithm based on granular computing

Aiming to resolve the problems of the traditional K-means clustering algorithm such as random selecting of initial clustering centers, lacking the ability of handling boundary objects of data, the low efficiency, and low clustering accuracy, this paper proposed a new K-means clustering algorithm. This algorithm firstly introduced in the granular computing theory and selected the initial centers based on the density and max-min distance means, so that the initial clustering center could be in the different class. Then the algorithm combined with rough set and adjusted dynamically the boundary objects by changing the weighting factors of upper approximation set and boundary set, to settle the clustering problem of boundary data. Finally, it used criterion function of equalization based on the within-class distance/the distance between classes as the judgment condition of algorithm termination, to get better clustering effect. The results of experiments show that this algorithm has higher accuracy, less iteration times, and reduces the sensitive degree of the noise.