Application of Dimension Reduction on Using Improved LLE Based on Clustering

Locally linear embedding (LLE) is one of the methods intended for dimension reduction. Its extension using clustering and improved LLE for dimension reduction is investigated. Firstly, using clustering can reduce time-consuming. Secondly, the improved LLE is suitable for selecting the number K of the nearest neighbors. When the number K of the nearest neighbors is small, it can obtain good results. While the original LLE algorithm obtains the same results, the number K of nearest neighbors may be much larger. Even if the number K of the nearest neighbors using the improved LLE is selected to be larger, the result is still right. So, the improved LLE is not sensitive to the selection of K. It is shown that the improved LLE based on clustering has less computing than the original LLE algorithm and enlarges the choice of parameter K by experiment.