| Â | PWM | PMM1 | PMM2 | PMM3 | PMM4 |
|---|
PWM | – | 0+0 | 0+0 | 0+0 | 0+0 |
PMM1 | 15+6 | – | 0+0 | 1+0 | 2+0 |
PMM2 | 15+8 | 4+6 | – | 0+0 | 2+0 |
PMM3 | 14+9 | 6+4 | 1+1 | – | 0+0 |
PMM4 | 16+9 | 7+5 | 3+3 | 2+0 | – |
- For each data and every combination of models, we perform a Wilcoxon signed rank test (α=0.01) comparing the distribution of AUC values from the different cross-validation iterations. The entry in a+b the i-th row and the j-th column denotes the number of data sets for which the model corresponding to row i yields a significantly better classification performance than the model corresponding to column j, where a denotes the number of significant differences in category A data sets, and b the number of significant differences in category b data sets. We find many instances where increasing the model order yields to significantly better classification (lower triangle), but only few instances where it yields a significantly worse classification (upper triangle)
