Abstract: In order to solve the problem of under-fitting state of LPP algorithm in practical application,in this paper, the mapping principle of Locality Preserving Projections (LPP) is discussed in detail. The relationship of LPP method between the under-fitting state on certain dataset and adjacency graph is analyzed. The LPP manifold learning method (ISOLPP) is proposed on the basis of geodesic. The experiment results show that the good embedded effect is achieved by implenmenting ISOLPP method on multiple test data sets. It significantly improves the adaptability of the algorithm by not only inheriting the advantages of LPP algorithm with explicit projection matrix, but also solving the disadvantages of LPP algorithm in the under-fitting state.

Key words: Locality Preserving Projections, under-fitting state, geodesic