Table 4. LOOCV accuracy (%) of classifiers.
| Method | Leukemia | CNS | DLBCL | Prostate1 | Prostate3 | Lung | GCM | Average |
| COSSY [KEGG] | 98.6 | 85.3 | 93.5 | 90.2 | 100.0 | 99.5 | 85.0 | 93.2 |
| COSSY [STRING] | 95.8 | 88.2 | 94.8 | 90.2 | 97.0 | 98.3 | 84.6 | 92.7 |
| DIRAC | 94.8 | 72.3 | 73.4 | 62.9 | 100.0 | 98.8 | 75.2 | 82.5 |
| k-TSP* | 95.8 | 97.1 | 97.4 | 91.2 | 97.0 | 98.9 | 85.4 | 94.7 |
| TSP* | 93.8 | 77.9 | 98.1 | 95.1 | 97.0 | 98.3 | 75.4 | 90.8 |
| SVM* | 98.6 | 82.4 | 97.4 | 91.2 | 100.0 | 99.5 | 93.2 | 94.6 |
| Doublet [Sign-DT]+ | 93.1 | 82.4 | 97.4 | 86.3 | 97.0 | 98.3 | 85.0 | 91.3 |
| Doublet [Sumdiff-DT]+ | 91.7 | 70.6 | 97.4 | 82.4 | 87.9 | 95.0 | 81.4 | 86.6 |
| Doublet [Mul-DT]+ | 84.7 | 55.9 | 97.4 | 86.3 | 90.9 | 92.3 | 83.2 | 84.4 |
| Decision Tree (DT)* | 73.6 | 67.7 | 80.5 | 87.3 | 84.9 | 96.1 | 77.9 | 81.1 |
| Nave Bayes* | 100.0 | 82.4 | 80.5 | 62.8 | 90.9 | 97.8 | 84.3 | 85.5 |
| k Nearest Neighbor* | 84.7 | 76.5 | 84.4 | 76.5 | 87.9 | 98.3 | 82.9 | 84.5 |
| PAM* | 97.2 | 82.4 | 85.7 | 91.2 | 100.0 | 99.5 | 79.3 | 90.7 |
The leftmost column contains the names of the methods; the rightmost column shows the average accuracy of each method for seven datasets, and other columns show the accuracy (%) for individual datasets. ‘COSSY [KEGG]’ and ‘COSSY [STRING]’ represent COSSY using KEGG and STRING, respectively. ‘DIRAC’ is the algorithm proposed in [17] whose LOOCV accuracies have been calculated using the matlab code published with the paper. k-TSP and TSP denote the classification algorithms described in [5] and [4], respectively. SVM stands for Support Vector Machine. ‘Doublet [Sign-DT]’, ‘Doublet [Sumdiff-DT]’, and ‘Doublet [Mul-DT]’ denote the classification methods using Sign-Doublet, Sumdiff-Doublet, and Mul-Doublet, respectively, with decision trees as described in [6]. The last three rows contain the loocv accuracies using Nave Bayes, k Nearest Neighbor, and PAM classifier, respectively.
Results obtained from [5].
Results obtained from [6].