Research Article
An Ad Hoc Random Initialization Deep Neural Network Architecture for Discriminating Malignant Breast Cancer Lesions in Mammographic Images
Table 2
Performance statistics for our best performing models as evaluated on the test set.
| | Model 1 (best AUC overall on tde validation set, point witd bestF1 score on tde test set) | | Accuracy | PPV (precision) | FDR | TPR (recall, sensitivity) | FNR (missrate) | FPR (fall out) | TN (specificity) | F1 score | F2 score | F5 score | | 71.19% | 59.80% | 40.20% | 84.40% | 15.60% | 37.56% | 62.44% | 70.00% | 77.98% | 63.50% |
| | Model 2 (best F2 score overall on the validation set, point with best F2 score on the test set) | | Accuracy | PPV (precision) | FDR | TPR (recall, sensitivity) | FNR (missrate) | FPR (fallout) | TN (specificity) | F1 score | F2 score | F5 score | | 55.93% | 47.40% | 52.60% | 97.16% | 2.84% | 71.36% | 28.64% | 63.72% | 80.30% | 52.81% |
|
|
