Table 5 Performance of classical ML classifiers and presented DNN model applied with the same preprocessing and 10-fold nested-CV evaluation. Values indicate mean ± SD (%) over ten outer folds. The results shows that DNN exhibits a better mean performance and less fluctuation, which shows that it possess good generalization with leakage-free evaluation.
Classifier | Sensitivity | Specificity | Precision | F1-Score | Accuracy | AUC-ROC |
|---|---|---|---|---|---|---|
SVM | 85.22 ± 0.84 | 79.75 ± 0.92 | 83.78 ± 1.01 | 84.50 ± 0.79 | 82.77 ± 0.85 | 0.8241 |
NB | 89.46 ± 0.71 | 82.48 ± 0.88 | 86.51 ± 0.73 | 87.96 ± 0.69 | 86.36 ± 0.72 | 0.9127 |
LR | 90.55 ± 0.62 | 88.24 ± 0.67 | 91.45 ± 0.59 | 91.00 ± 0.56 | 89.58 ± 0.60 | 0.9302 |
RF | 90.44 ± 0.66 | 85.53 ± 0.73 | 88.63 ± 0.68 | 89.53 ± 0.63 | 88.26 ± 0.65 | 0.9134 |
KNN | 83.51 ± 0.91 | 81.01 ± 0.87 | 84.38 ± 0.88 | 83.94 ± 0.85 | 82.39 ± 0.84 | 0.8773 |
DT | 88.38 ± 0.83 | 78.28 ± 0.97 | 82.57 ± 0.90 | 85.37 ± 0.81 | 83.71 ± 0.86 | 0.8809 |
ANN | 93.05 ± 0.53 | 92.04 ± 0.56 | 93.98 ± 0.49 | 93.51 ± 0.47 | 92.61 ± 0.52 | 0.9475 |
Proposed | 94.06 ± 0.52 | 94.67 ± 0.45 | 95.96 ± 0.37 | 95.00 ± 0.33 | 94.32 ± 0.41 | 0.9623 |
