基于模糊聚类和测地距离的 LLE 算法

摘要

Locally linear embedding (LLE) algorithm is a method to solve the problem of dimension reduction .To calcu‐late weight matrix and determine the quantity of neighboring points ,this paper proposes LLE algorithm which is based on fuzzy clustering and geodesic distance .Fuzzy C means clustering can reduce the amount of calculation of the weight value matrix and computing time .The application of the LLE algorithm works very well when a small number of neighboring points are chosen .Experimental results show certain advantages of the LLE algorithm as it greatly reduces the calculation M matrix and neighboring point calculation .%局部线性嵌入算法（LLE）是一种解决降维问题的方法，针对权值矩阵的计算及近邻点个数选取，提出了基于模糊聚类和测地距离的LLE算法，模糊C均值聚类可以减少计算权值矩阵的计算量，缩减计算时间；使用测地距离的LLE算法可以在选取近邻点个数较小的情况下获得良好的效果。实验结果表明，基于模糊C均值聚类和测地距离的LLE算法大大缩减了计算M矩阵和近邻点的计算量，具有一定的优越性。