LOCATION:   Snedecor 319
                                                
SPEAKER:   Adele Cutler, Department of Mathematics and Statistics,

                                           Utah State University, Logan, Utah
 

TITLE: Random Forests - a Look Inside the "Black Box"
 

ABSTRACT:

Random Forests are among the most accurate off-the-shelf classifiers available, and they can also be used for cluster analysis and regression.  They do not overfit, so they can handle problems in which there are thousands of variables and relatively few observations. Examples include microarray and mass spectrometry analysis. However, statisticians want more than accurate prediction: we usually want to understand what is going on in the problem, and answers to such questions as:     * how well separated are the groups?     * are there unusual sub-groups?     * which variables are related to the groups, and how?   In this talk, I will summarize the random forests algorithm and demonstrate some visualization software that allows us to answer these questions (and more), using examples from microarray analysis, the genetics of autism, and satellite imaging. * - Joint work with Leo Breiman, Department of Statistics, University of California, Berkeley COFFEE: 3:45 p.m., 104 Snedecor Hall