Application Algorithms of Nonlinear Dimensionality Reduction to Material Database Visualization
Олексій Водка, Ivan Zadorozhniy, Ruslan Lavshhenko · 2020 IEEE 15th International Conference on Computer Sciences and Information Technologies (CSIT) · 2020
In our world there is a huge quantity of different materials, alloys, compositions, substances. They have different mechanical properties, chemical compositions, different colors, smells, and, most importantly, applications in human life, etc. Because of so many properties of various compounds, it poses the problem of storing information about them and also creates a space for their visualization in different ways.This work deals with the creation of an online application and database that makes it possible to use information about metals, their alloys, and the search for the closest analogs. The web application gives the user a wide selection of devices to use and also uninstalls additional software products. During development, we used the general-purpose programming language Python with Flask library and web technologies HTML, CSS, and JavaScript. For the correct search for similar materials, a reasonable search is developed and implemented in an online application. The database is created based on SQL technologies the database management system - MySQL. To obtain information about metals on the Internet, a parser program is developed that selectively received data from open sources and recorded them in the created database.Humans can effectively use visual information in a maximum of three projections. Therefore, the task of reducing the dimension of information, and its visualization is one of the main directions when working with multidimensional data. The solution to this problem allows the researcher to correctly build assumptions and manipulate data in the course of reasoning or evidence.The task of the correct visualization of multidimensional data on the plane and three-dimensional space is solved by several well-known algorithms, namely: Isomap, t-SNE, MDS. As a result of the visualization of all the collected information about metals, clusters of materials is identified that correspond to the already known classes of metals. The study confirmed the correctness of the algorithms for reducing the dimension, the correctness of the information collected, the usability of the web application.