A new measure of classifier diversity in multiple classifier system

Diversity among the team has been recognized as a very important characteristic in classifier combination. There are varied diversity measures. They can be categorized into two types, pairwise diversity measures and non-pairwise diversity measures. Above diversity measures are defined based on Oracle outputs of classifier. While using diversity measures to calculate diversity of classifiers that have soft label outputs, much information about class will be lost. That is a weakness of above measures. In order to solve the problem, this paper puts forward a new diversity measure, which can be used in the classifiers that have soft label outputs. Experimental results show that it contains more information about classifier outputs and accurately reflects the difference of classifier outputs.