Other Statistics and Probability Commons^™

Rfviz: An Interactive Visualization Package For Random Forests In R, Christopher Beckett

All Graduate Plan B and other Reports, Spring 1920 to Spring 2023

Random forests are very popular tools for predictive analysis and data science. They work for both classification (where there is a categorical response variable) and regression (where the response is continuous). Random forests provide proximities, and both local and global measures of variable importance. However, these quantities require special tools to be effectively used to interpret the forest. Rfviz is a sophisticated interactive visualization package and toolkit in R, specially designed for interpreting the results of a random forest in a user-friendly way. Rfviz uses a recently developed R package (loon) from the Comprehensive R Archive Network (CRAN) to create …

Imputation For Random Forests, Joshua Young

All Graduate Plan B and other Reports, Spring 1920 to Spring 2023

This project introduces two new methods for imputation of missing data in random forests. The new methods are compared against other frequently used imputation methods, including those used in the randomForest package in R. To test the effectiveness of these methods, missing data are imputed into datasets that contain two missing data mechanisms including missing at random and missing completely at random. After imputation, random forests are run on the data and accuracies for the predictions are obtained. Speed is an important aspect in computing; the speeds for all the tested methods are also compared.

One of the new methods …

Tutorial For Using The Center For High Performance Computing At The University Of Utah And An Example Using Random Forest, Stephen Barton

All Graduate Plan B and other Reports, Spring 1920 to Spring 2023

Random Forests are very memory intensive machine learning algorithms and most computers would fail at building models from datasets with millions of observations. Using the Center for High Performance Computing (CHPC) at the University of Utah and an airline on-time arrival dataset with 7 million observations from the U.S. Department of Transportation Bureau of Transportation Statistics we built 316 models by adjusting the depth of the trees and randomness of each forest and compared the accuracy and time each took. Using this dataset we discovered that substantial restrictions to the size of trees, observations allowed for each tree, and variables …