methods such as prototyping and (parallel) algorithmic design has been used. We
applied statistical methods (briefly described below) for quantitative analysis.
empirical results we have applied statistical hypothesis tests. The most frequently
used have been the t-test together with output from 10-fold cross validation
experiments in order to compare the average value (in our work: classification
accuracy and computational performance), Cohen [1995]. In paper I where we
compared computational performance and classification accuracy between our
proposed classifier with other classifiers on several datasets, we used pairwise
t-tests on the experimental output from all applicable datasets in order to test
how the classifiers compared overall, Lange [1999].
ulator interacted and controlled the computational performance we used a full
factorial design, Montgomery [1997]. The result from the factorial design is shown
in paper D.