A research based on application of dimension reduction technology in data visualization using machine learning

At present, the research work of dimension reduction at home and abroad is more about the theoretical exploration and application research of specific dimension reduction methods, and the research on the basic theory of dimension reduction and the spatial properties of high-dimensional data is not perfect. To solve the problem of the application of dimension reduction technology optimization in data visualization, a method of ISOMAP algorithm using Artificial Intelligence is proposed. The specific content of this method uses nonlinear Sammon mapping with short edge first preservation, to improve the "over-clustering" phenomenon of ISOMAP algorithm. When visualizing such data using Artificial Intelligence techniques, furthermore, the application scope of ISOMAP algorithm is extended to a certain extent. The experimental results show that the nonlinear Sammon mapping is sensitive to the initial value and the learning rate, to this end, we use the classical MDs (Molecular dynamics) algorithm to provide a relatively good initial value for the Sammon mapping, we choose three different neighborhood sizes: K = 8, K = 12 and K = 16. K = 8 is appropriate, while K = 12 and K = 16 are inappropriate, this can be verified by the corresponding residual, and the variable substitution method without learning rate is adopted to replace the gradient descent method which is more sensitive to the learning rate. It is proved that the application of nonlinear dimension reduction technology optimization in data visualization has been greatly improved.