Massive data visualization based on dimensionality reduction and projection error evaluation

In thesis two ways to evaluate projection error for massive data sets are proposed. One of them is based on building the sample of the data set, the second one on dividing the data set into the smaller data sets. Both proposed ways of projection error evaluation are suitable for massive data sets and allow us to decrease computation time as well as to reduce the usage of computer operating memory. In this dissertation new approach of massive data visualization was proposed. The proposed visualization approach consist of two stages: selection of data subset; visualization of the projection of the data subset. This approach allows us to visualize data without points overlapping and keeps the structure of the data. All proposed approaches in this dissertation were applied to solve real world data tasks. Comprehensive analysis of various dimensionality reduction techniques was performed while solving the dimensionality reduction problem. Analysis included various classic dimensionality methods and methods which are based on control point’s selection. The main results of the dissertation were published in 6 research papers: 3 papers are published in periodicals, reviewed scientific journals and 3 papers are published in conference proceedings. The main results have been presented and discussed at 3 national and 3 international conferences. The dissertation consists of 6 chapters and the list of references The scope of the work is 119 pages including 25 figures and 19 tables. The list of references consists of 89 sources