Local linear embedding algorithms based on probability distributions

In response to the problem that the local linear embedding algorithm (LLE) only uses the concept of nearest neighbour to binarize the assignment of sample relations (there are only two fixed value cases of 0 and 1) and cannot judge or distinguish the degree of difference between any two points on the manifold, and the problem that the LLE algorithm does not make full use of the discriminative information of the data, this paper attempts to combine the discriminative information of the data and proposes a kernel based on probability distribution density estimation local linear embedding algorithm (KDELLE) in combination with discriminative information of the data. The algorithm obtains a natural one by estimating the kernel density of the sample on the manifold, which can be used to represent the affiliation score of a point on the manifold to a certain class of points, and adds discriminative information to obtain a weight matrix based on the probability distribution to further represent the distribution of points on the manifold. Through experimental comparative analysis on the bearing dataset, the method not only has a good clustering effect but also has good feature extraction results.