Table 3 Best-performing models with 946 interlandmark distances.
Classifier and best tuning parameters | Accuracy | Sensitivity | Specificity | PPV | NPV | Selected Features |
|---|---|---|---|---|---|---|
LR with excluded region-specific variables {scaled, ‘C’: 0.01, ‘penalty’: ‘l2’, ‘solver’: ‘newton-cg’} | ||||||
CV | 0.90 | 0.89 | 0.91 | 0.91 | 0.89 | without feature selection, n (features) = 208 |
Test | 0.95 | 0.95 | 0.95 | 0.95 | 0.95 | |
Test ST | 0.98 | 1.00 | 0.95 | 0.95 | 1.00 | |
Test ZG | 0.93 | 0.90 | 0.95 | 0.95 | 0.90 | |
LDA with excluded correlated variables {scaled, ‘shrinkage’: ‘auto’, ‘solver’: ‘lsqr’} | ||||||
CV | 0.94 | 0.93 | 0.95 | 0.95 | 0.93 | RFE, n (features) = 99 |
Test | 0.96 | 0.98 | 0.95 | 0.95 | 0.97 | |
Test ST | 0.95 | 1.00 | 0.90 | 0.91 | 1.00 | |
Test ZG | 0.98 | 0.95 | 1.00 | 1.00 | 0.95 | |
LDA excluded region-specific variables {‘shrinkage’: 0.1, ‘solver’: ‘lsqr’} | ||||||
CV | 0.93 | 0.93 | 0.92 | 0.92 | 0.93 | RFE, n (features) = 23 |
Test | 0.95 | 0.98 | 0.93 | 0.93 | 0.97 | |
Test ST | 0.98 | 1.00 | 0.95 | 0.95 | 1.00 | |
Test ZG | 0.93 | 0.95 | 0.90 | 0.90 | 0.95 | |
