Improvement And Implementation Of Manifold Learning Algorithms ISOMAP

Dimension reduction is an effective means to deal with these data. It can not only avoid "dimension disaster" effectively, but also find the inherent regularity of the data. The aim is:in premise of not change essence of original high-dimensional data structures, reduce or remove redundant information as far as possible and reduce the dimensions of original data, so as to achieve the purpose of dimension reduction. Through existing linear dimension reduction methods can gain true geometry of linear high-dimensional data.As real-world data are mostly non-linear, we need method to deal with non-linear high dimensional data effectively, however traditional linear dimension reduction decide to linear nature, such methods can only be used to find global linear structure of high-dimensional data, can not identify non-linear structure of high-dimensional effectively. Manifold learning methods have emerged in this context, to solve the problems of analysising nonlinear high-dimensional data, can effectively find the inside geometry of the nonlinear high-dimensional data.Isometric feature mapping algorithm is a global optimization manifold learning method, which the embedded results may reflect the manifold distance of high dimensional data, can gain the ideal embedded result. An important issue of the algorithm is more computation time. In order to reduce the computational time of Isometric feature mapping algorithm.To address this issue, this paper proposes using fuzzy c-means Clusteringing method select landmarks, to improve Isomap algorithm. Firstly, use fuzzy Clusteringing algorithm categorizing for high dimensional data sets. Secondly, find out Clustering Cens and as landmark of isometric mapping algorithm. Finally, solve the final embedding result by LMDS.In addition, whether ISOMAP algorithm gain embedding result from high dimensional data or not depends on the number of neighborhood, how to select the appropriate points in the neighborhood is still an open question. With theory of FCM and graph, this paper proposed tentative neighborhood value estimated algorithm, to determine ISOMAP algorithm parameters-the neighborhood value.