Table 4.
The best classification performances for predicting the BCR within five years after first diagnosis.
| Classifier | Feature set | AUC (%) | Accuracy (%) | Sensitivity (%) | Specificity (%) |
|---|---|---|---|---|---|
| (# selected features) | Mean ± se | Mean ± se | Mean ± se | Mean ± se | |
| Naive Bayesian | NCA (12) | 74.72 ± 0.35 | 72.14 ± 1.13 | 70.98 ± 1.94 | 72.50 ± 1.91 |
| RF (8) | 81.34 ± 0.24 | 74.68 ± 1.50 | 74.15 ± 2.77 | 74.85 ± 2.78 | |
| SVM-RFE (13) | 75.61 ± 0.42 | 72.60 ± 1.12 | 76.34 ± 1.82 | 71.44 ± 1.77 | |
| RF | NCA (10) | 88.25 ± 0.21 | 75.55 ± 0.68 | 90.44 ± 0.92 | 70.30 ± 1.10 |
| RF (3) | 90.36 ± 0.21 | 77.05 ± 0.94 | 94.15 ± 1.59 | 71.74 ± 1.70 | |
| SVM-RFE (6) | 89.14 ± 0.25 | 78.84 ± 1.10 | 86.59 ± 2.03 | 76.44 ± 2.05 | |
| SVM | NCA (5) | 80.69 ± 0.61 | 84.22 ± 0.69 | 70.00 ± 1.46 | 88.64 ± 1.11 |
| RF (6) | 82.89 ± 0.57 | 85.84 ± 0.64 | 73.90 ± 0.89 | 89.55 ± 0.92 | |
| SVM-RFE (5) | 80.10 ± 0.48 | 83.18 ± 0.91 | 67.56 ± 2.70 | 88.03 ± 1.97 |
Mean and standard error (se) of AUC values, Accuracy, Sensitivity, and Specificity evaluated on 100 10-fold cross-validation rounds for each feature importance technique and classifier.