结合引力的模糊C-值聚类算法研究

RESEARCH ON FUZZY C-MEANS CLUSTERING ALGORITHM COMBINING GRAVITY

基于距离的模糊聚类算法是把数据对象视为互相独立的,虽然在一定程度上反映了数据对象间的位置关系,但不能反映多重关系,使计算量急剧增加,时间复杂度高,收敛速度慢。对模糊C-均值聚类算法进行了改进,在原有的模糊C-均值聚类算法基础上,引入了物理学中的万有引力思想,提出了一种基于引力改进的模糊聚类算法。实验分析表明,该算法能够较好地克服基于距离的模糊聚类算法仅考虑单一位置关系的缺点,并且在一定程度上降低了时间复杂度,提高了算法的收敛速度,聚类效果较好。

Distance-based fuzzy clustering algorithm regards data objects as independent objects each other. Although this reflects the loca- tion relationship among data objects to some extent, but it can not reflect multiple relationships, moreover, it causes dramatic increase in computation load, high time complexity and slow convergence rate. This paper proposes an improvement on fuzzy C-mean clustering algorithm, the gravity improvement-based fuzzy clustering algorithm is presented by introducing the idea of universal gravity law in physics on the basis of previous fuzzy C-means clustering algorithms. The analysis on experimental result shows that this algorithm can be used to well overcome the shortcoming of distance-based fuzzy clustering algorithm which only takes single position relation into account. Meanwhile it reduces the time complexity to some degree, improves the algorithm＇ s convergence rate and achieves better clustering effect.