Salon 2
Improvements in Random Forests and Detecting Interactions (303251)
Adele Cutler, Utah State University*Anna Quach, Utah State University
Keywords: Epistatic Interactions, Random Forests, Visualizing Asymmetric Proximities
Random Forests was developed in 2001 by Leo Breiman. Since then there have been 20,251 (2/16/2016) citations according to Google Scholar. Random Forests has been used successfully in a wide variety of problems. Over the past couple of years, I've been working with Dr. Adele Cutler, co-developer of Random Forests, to find ways to improve Random Forests in particular situations. In this presentation, I'll be talking about the problems I've been working on for my PhD.
A major part of my dissertation concerns detecting SNP-SNP interactions. For complex diseases, it is believed that interactions between SNPs may be associated with the disease, even when no individual SNP is strongly associated. Random Forests is known to automatically detect such interactions in other contexts. However, when the number of variables is very large, variables with interactions but no main effect are not easily detected by Random Forests. I will introduce a new feature screening method that ranks interactions to improve the performance of Random Forests, and a new asymmetric model to aid in interpreting the results.