Table 3.
Comparison of the average predictive performance by k-fold cross-validation with optimal-feature subset
| LR (95% CI) | KNN (95% CI) | NB (95% CI) | SVM (95% CI) | P | |
|---|---|---|---|---|---|
| Sen | 88.1% (81.7%–94.5%) | 83.2% (75.7%–90.6%) | 90.1% (84.2%–96.0%) | 85.1% (78.1%–92.2%) | 0.479 |
| Spe | 88.7% (83.9%–93.5%) | 93.5% (89.7%–97.2%)a | 84.5% (79.0%–90.0%)b, c | 94.0% (90.4%–97.7%)a | 0.01 |
| FPR | 11.3% (6.5%–16.1%) | 6.5% (2.8%–10.3%)a | 15.5% (10.0%–21.0%)b, c | 6.0% (2.3%–9.6%)a | 0.01 |
| FNR | 11.9% (5.5%–18.3%) | 16.8% (9.4%–24.3%) | 9.9% (4.0%–15.8%) | 14.9% (7.8%–21.9%) | 0.479 |
| PPV | 82.4% (75.1%–89.7%) | 88.4% (81.9%–95.0%)a | 77.8% (70.1%–85.4%)b,c | 89.6% (83.4%–95.8%)a | 0.065 |
| NPV | 92.5% (88.4%–96.6%) | 90.2% (85.8%–94.7%) | 93.4% (89.4%–97.4%) | 91.3% (87.1%–95.6%) | 0.736 |
| Accuracy | 88.5% (84.6%–92.3%) | 89.6% (85.9%–93.3%) | 86.6% (82.5%–90.7%) | 90.7% (87.2%–94.2%) | 0.48 |
| F1 score | 0.8517 | 0.8517 | 0.8349 | 0.8731 | — |
| AUC | 0.937 (0.902–0.972) | 0.949 (0.924–0.973) | 0.935 (0.906–0.964) | 0.931 (0.895–0.967) | — |
Notes: aP<0.05 vs NB; bP<0.05 vs KNN; cP<0.05 vs SVM. P-values denote overall statistical results for the four models.