Validation of voting committees

This article contains a method to bound the test errors of voting committees with members chosen from a pool of trained classifiers. There are so many prospective committees that validating them directly does not achieve useful error bounds. Because there are fewer classifiers than prospective committees, it is better to validate the classifiers individually than use linear programming to infer committee error bounds. We test the method using credit card data. Also, we extend the method to infer bounds for classifiers in general.