Table 3.
Comparison of the predictive performance of different models in optimal feature subset in test set.
| Variable | LR (95%CI) | QDA (95%CI) | NB (95%CI) | SVM (95%CI) | Adaboost (95%CI) | BP (95%CI) | p value |
|---|---|---|---|---|---|---|---|
| SEN | 58.54% (42.20%-73.30%) | 60.98% (44.54%-75.38%) | 73.17% (56.69%-85.25%) | 60.98% (44.54%-75.38%) | 80.49% (64.63%-90.63%) | 75.61% (59.36%-87.09%) | 0.15 |
| SPE | 93.33% (84.47%-95.52%)a,c,d,e | 86.67% (76.39%-93.08%)d | 76.00% (64.50%-84.79%)b | 89.33% (79.54%-94.95%)d | 73.33% (61.66%-82.58%)a,b,f | 74.67% (63.08%-83.69%)a,b | 0.001 |
| FPR | 6.67% (1.02%-12.32%)a,c,d,e | 13.33% (5.64%-21.03%) d | 24.00% (14.33%-33.67%) b | 10.67% (3.68%-17.66%) d | 26.67% (16.66%-36.67%) a,b,f | 25.33% (15.49%-35.17%) a,b | 0.001 |
| FNR | 41.46% (26.38%-56.54%) | 39.02% (24.09%-77.80%) | 26.83% (13.27%-40.39%) | 39.02% (24.09%-77.80%) | 19.51% (7.38%-31.64%) | 24.39% (11.25%-37.53%) | 0.15 |
| PPV | 82.76% (63.51%-93.47%) | 71.43% (53.48%-84.76%) | 62.50% (47.33%-75.68%) | 75.76% (57.37%-88.26%) | 62.26% (47.87%-74.88%) | 62.00% (47.16%-75.00%) | 0.281 |
| NPV | 93.33% (84.47%-97.52%) | 80.25% (69.61%-87.95%) | 83.82% (72.47%-91.27%) | 80.72% (70.29%-88.25%) | 87.30% (75.96%-93.97%) | 84.85% (73.44%-92.11%) | 0.87 |
| Accuracy | 80.3% (73.0-87.7%) | 78.5% (71.1-85.9%) | 75.0% (67.0-83.0%) | 79.3% (71.8-86.8%) | 75.9% (68.0-83.8%) | 75.0% (67.0-83.0%) | 0.831 |
| AUC | 0.782 (0.694-0.853) | 0.785 (0.686-0.848) | 0.779 (0.688-0.849) | 0.772 (0.679-0.842) | 0.826 (0.740-0.888) | 0.805 (0.714-0.869) | / |
aCompared with QDA, p < 0.05; bCompared with LR, p < 0.05; cCompared with NB, p < 0.05; dCompared with AdaBoost, p < 0.05; eCompared with BP, p < 0.05; fCompared with SVM, p < 0.05. p value denoted the overall statistical result for the four models.