For classification of low-dimensional data is very common, but not for the classification of high-dimensional data, mainly because of too high dimension. In particular, for the uneven distribution of the sample set, the traditional locally linear embedding(LLE) algorithm is vulnerable to the impact of the number of nearest neighbor points, In order to overcome this problem, this paper improves locally linear embedding algorithm by changing the distance. Through the experiments indicates that the improved distance locally linear embedding algorithm can make the original sample set distribute evenly as far as possible, thereby reducing the influence of selection of the number of nearest neighbor points on locally linear embedding, on the premise of ensuring accurate classification, to achieve the purpose of effectively shorten the time.