Title: Bayesian probit classification trees
Authors: Paola Stolfi - CNR - Institute for Applied Mathematics (Italy) [presenting]
Mauro Bernardi - University of Padova (Italy)
Daniele Durante - University of Padova (Italy)
Abstract: Ensemble of decision trees are popular techniques for regression and classification either because of their forecasting performances and their ability to account for complex nonlinear dependence structures among predictors. Leveraging on the Bayesian Additive Regression Trees (BART) approach, we propose new methods to deal with binary classification for CART and BART. Specifically, we introduce a new representation for the probit classification model that avoid the data augmentation scheme previously used. The proposed approach is illustrated and validated through comparison with alternative methods on simulated and real datasets.