"Induction of Survival Trees by Quadratic Splits and Dipolar Splitting Criteria," Dr. Drew Lazar, September 29, 1:00PM, RB 125.
Abstract: Ensemble methods in Survival Analysis depend on splitting data at nodes in underlying decision trees. Various splitting criteria have been proposed and implemented using within- or between-node homogeneity. Criteria in the former category include using log-likelihood statistics based on parametric assumptions and criteria in the latter category often depend on the log-rank statistic. We improve and clarify existing algorithms which rely on non-parametric dipolar splits by hyperplanes for maximizing between-node homogeneity. We demonstrate improved prediction of survival experience and more parsimonious survival trees using simulated and real data sets. We extend these methods to non-linear surfaces while avoiding overfitting and reducing tree sizes while maintaining predictive power.