Meta-classifiers and selective superiority

Abstract

Given that no one classification method is the best in all tasks, a variety of approaches have evolved to prevent poor performance due to mismatch of capabilities. One approach to overcome this problem is to determine when a method may be appropriate for a given problem. A second, more popular approach is to combine the capabilities of two or more classification methods. This paper provides some evidence that the combining of classifiers can yield more robust solutions.