In discrete discriminant analysis (DDA) different models often exhibit different classification performances. Therefore, the idea of combining models has increasingly gained importance. In the present work we focus on the evaluation of alternative DDA models, including combined models. The proposed approach uses not only the classic indicators of classification precision but also indices of agreement that regard the relationship between the actual classes and the ones predicted by discriminant analysis. The performance of the DDA methods is analyzed based on simulated binary data, using small and moderate sample sizes. The results obtained illustrate the potential of combining DDA models, offering different evaluation perspectives.

--

Adjust Rand index,Classification precision,Affinity coefficient,Simple agreement,Class conditional probability

• Mathematics - Natural Sciences