Model switching for bayesian classification trees with soft splits

Abstract

Due to the high number of insolvencies in the credit business, automatic procedures for testing the credit-worthiness of enterprises become increasingly important. For this task we use classification trees with soft splits which assign the observations near the split boundary to both branches. Tree models involve an extra complication as the number of parameters varies as the tree grows and shrinks. Hence we adapt the reversible jump Markov Chain Monte Carlo procedure to this model which produces an ensemble of trees representing the posterior distribution. For a real-world credit-scoring application our algorithm yields lower classification errors than bootstrapped versions of regression trees (CART), neural networks, and adaptive splines (MARS). The predictive distribution allows to assess the certainty of credit decisions for new cases and guides the collection of additional information.