Table 9. Rank differences and p-values for pair-wise comparison of classification algorithms.
| Hypothesis | Rank diff | p-value |
|---|---|---|
| RBF vs LVQ | 4.07 | < 10−15 |
| Lin vs LVQ | 3.62 | 2.465 ⋅ 10−13 |
| RBF vs DT | 3.56 | 6.461 ⋅ 10−13 |
| MLP vs LVQ | 3.52 | 1.159 ⋅ 10−12 |
| Lin vs DT | 3.11 | 6.542 ⋅ 10−10 |
| RF vs LVQ | 3.06 | 1.286 ⋅ 10−9 |
| MLP vs DT | 3.01 | 2.501 ⋅ 10−9 |
| RF vs DT | 2.55 | 7.156 ⋅ 10−7 |
| KNN vs LVQ | 2.38 | 4.841 ⋅ 10−6 |
| KNN vs DT | 1.87 | 0.0008079 |
| RBF vs KNN | 1.69 | 0.003693 |
| Lin vs KNN | 1.24 | 0.08629 |
| MLP vs KNN | 1.14 | 0.138 |
| RBF vs RF | 1.01 | 0.2228 |
| RF vs KNN | 0.68 | 1 |
| Lin vs RF | 0.56 | 1 |
| RBF vs MLP | 0.55 | 1 |
| DT vs LVQ | 0.51 | 1 |
| MLP vs RF | 0.46 | 1 |
| RBF vs Lin | 0.45 | 1 |
| Lin vs MLP | 0.10 | 1 |
For each pair-wise comparison of classification algorithms, the rank difference was calculated as the difference of the average ranks over all data sets. The first algorithm of each hypothesis is the better one (lower rank). p-values were corrected using Shaffer’s static method.