Study And Application Of Several Improved Methods Of Nonlinear Dimension Reduction For High Dimensional Data

High-dimensional data is an important product of high-tech science and technology,cigarette product quality and materials cost of production,Due to advances in science and technology,data collection becomes easier,such as biological genetic data,market economic data,text document data,digital image data,etc.The huge value of these data has gradually been recognized.However,it is very difficult to directly process high-dimensional data.As a very important data preprocessing method for data mining to deal with massive high-dimensional data,nonlinear dimensionality reduction technology can effectively convert high-dimensional data into a more compact low-dimensional representation.Thereby getting a meaningful low-dimensional data structure hidden in the high-dimensional data set,and extracting the data backbone information.Therefore,it is of great theoretical and practical application value to study efficient and accurate nonlinear dimensionality reduction methods for high dimensional data.In this paper,while mainly participating in the theoretical research of nonlinear dimensionality reduction improvement methods for high-dimensional data,carrying out the research work in the following three aspects with face image data.Firstly,aiming at the shortcomings of local linear embedding algorithm that Euclidean distance is not suitable for the search of nonlinear high-dimensional data neighbors,the local linear embedding algorithm based on Geodesic Rank-order distance metric is proposed based on the characteristics of geodesic distance and Rank-order distance.The superiority of the algorithm is verified by the comparison experiment between ORL face database and Yale face database.Secondly,using the class label information of some samples to readjust the similarity measurement method,a semi-supervised local linear embedding algorithm based on geodesic distance and class label information is proposed,and the low-dimensional coordinate information of the test set samples is obtained by combining the extreme learning machine,and the improved algorithm is proved to be a semi-supervised dimensions reduction method with better dimensions-reduction effect by experiments.Thirdly,the semi-supervised local linear embedding algorithm based on geodesic distance is used to reduce the dimension of the training set to obtain its low-dimensional coordinates as a priori information.It is used to guide the locallinear embedding algorithm to reduce the dimension.A semi-supervised local linear embedding algorithm based on geodesic distance and low-dimensional coordinate information is proposed.Compared with the above two improved algorithms in ORL face image and Yale face database,it is found that the semi-supervised local linear embedding algorithm based on low-dimensional coordinate information is more computationally complex,but has stronger classification performance.