Bayesian posterior misclassification error risk distributions for ensemble classifiers

Computing risk-based misclassification error density distribution for ensembles is an important yet difficult task. Bayesian methods provide one way to estimate these density distributions. In this paper, Bayesian modeling approach is used to compute posterior misclassification error density distributions for both binary and non-binary classifiers. Real-world datasets and holdout samples are used to illustrate computation of posterior misclassification error distributions. These posterior error distributions are very useful to compare ensembles, and provide risk-based misclassification cost estimates.