A Novel Recurrent Neural Net based Dimension Reduction Technique for Effectual Data Visualization

t-SNE, UMAP, & Tramp have shown excellent visualization performance on various real-world datasets. These approaches may preserve global structure but not local structure, a trade-off that has always plagued them. Computer science RNN machine learning is evolving. RNN models can reliably forecast numeric values or nominal classifications, but a general-purpose mechanism for RNN architecture construction has yet to be devised. Computer scientists must use trial-and-error to create accurate RNN models. Models often need multiple input parameters. When there are many input factors, it might be hard to determine which ones affect model prediction the most. Science calls this the "curse of dimensionality." Problems may be reduced to two dimensions, however. Humans can plan and understand two-dimensional problems. Visualizing a multi-dimensional dataset may reveal which input factors most affect a nominal or numeric outcome. Identifying these factors may help train RNN models faster by utilizing only input variables that impact the result variable. This study explores RNN training methods and uses dimensional reduction to see and interpret complicated data.