MBACT- Multiclass Bayesian Additive Classification Trees

In this article, we propose Multiclass Bayesian Additive Classifi-cation Trees (MBACT) as a nonparametric procedure to deal with multiclass classification problems. MBACT is a multiclass extension of BART: Bayesian Additive Regression Trees [Chipman et al., 2010]. In a range of data generating schemes and real data applications, MBACT is shown to have good predictive performance, competitive to existing procedures, and in particular it outperforms most proce-dures when the relationship between the response and predictors is nonlinear.