The Density-Based Descending Dimension Algorithm LLE
Danqiu Fu · 2019 4th International Conference on Mechanical, Control and Computer Engineering (ICMCCE) · 2019
When computer scientists expect to obtain valuable information through data mining in massive data, they are often confronted with the excessive feature of big data and complex high-dimensional structure. Therefore, researchers have introduced the concept of manifold learning to remove the redundant features which account for a large proportion but have no value, which can help improve learning efficiency by reducing computational complexity. In the actual process, however, there will always be discordant factors interfering with our judgment on the data. Therefore, this paper proposes a new, improved LLE dimension reduction algorithm based on data density judgment, which is simply referred to as DLLE. This algorithm is aimed to avoid the interference of noise points and the influence of surface boundary blur on dimension reduction caused by irregular manifold structure, thus better preserving the true geodesic distance of the data after the dimension reduction. An experiment is carried out on several sets of 3D synthetic datasets, whose results show that the improved algorithm is superior to the traditional LLE algorithm.