Results for the combined classifiers on training and test set.

The table summarizes the results achieved by the different combination approaches on the training set (10-fold cross validation; standard deviation in brackets) and the test set. The reference methods are Single Best and Oracle, the non-trainable combiners are named according to their function, the meta-classifiers according to the statistical learning methods. Decision Templates, Dampster-Shafer, Clustering and Local Accuracy are the methods described in detail in the Methods section. Feature is the combination on the feature level.