Online Program

Saturday, October 22
Knowledge
Community
Influence
Sat, Oct 22, 4:30 PM - 5:15 PM
Carolina Ballroom
Poster Session 6

Automatic Constrained Tree: An Approach to Accommodate Nested Data (303500)

*Rebecca R. Carter, Case Western Reserve University 

"Tree based modeling is extremely flexible for modeling complex and diversified populations, as well as for finding important features or covariates that impact the outcome. The covariates used to build a tree are often treated as independent to each other. In practice, however, one or more variables could naturally be nested inside the other, or the information of a particular covariate could only be collected if another covariate was equal to a certain value. An ad hoc solution for incorporating this dependent information is to manually redefine the covariates

In this work we develop a new tree-based modeling strategy, automatic constrained tree (acTree), which permits use of specified knowledge about the dependency of those specific variables. This allows for an optimal partition for building the model in the practical permissible domain. The performance of the acTree will be compared with several benchmark trees, and applied to Tibetan and Ovarian cancer studies. "