A valid diagnosis of migraine is a non-trivial decision problem. This is due to the fact that migraine can manifest wide range of varied symptoms. Thus, designing a computer aided diagnosis system for that problem remains still a very interesting topic. In this paper we present an ensemble classifier system designed for headache diagnosis. We assumed that the system should make fast initial diagnosis based on an analysis of data collected in the questionnaire only. Such an assumption eliminated possibility of application of most classical classification algorithms as they could not obtain decent level of accuracy. Therefore, we decided to apply an ensemble solution. Although it is clear that ensemble should consists of complementary classifiers, there is no guidance on how to choose ensemble size and ensure its diversity. Thus, we applied two stages strategy. Firstly, large pool of elementary classifiers were prepared. Its diversity was ensured by selecting algorithms of different types, structures, and learning algorithms. Secondly, we determined optimal size of the ensemble and selected its constituents using exhaustive search approaches. Results of experiments, which were carried on dataset collected in University of Novi Sad, shows that proposed system significantly outperformed all classical methods. Additionally we present analysis of diversity and accuracy correlation for tested systems.
ensemble classifier systems medical diagnosis support system headache diagnosis