An experimental study of classifier filtering

In classifier combination, some component classifiers may give wrong information. If the classification results of these classifiers are adopted by the combination algorithm, the final output will be wrong. So eliminating the component classifiers which give wrong information may improve the performance of the classifier combination algorithm. To distinguish this kind of algorithm, we call it classifier filtering. This paper presents an experimental study of classifier filtering. The experimental results on biometric data set show that classifier filtering method may improve the accuracy of the classifier combination algorithm effectively.