Skip to main content
. 2022 Jan 7;12:37. doi: 10.1038/s41598-021-03914-4

Table 2.

Performance characteristics of models in the validation set for predicting clinically relevant AHREs in patients with pacemaker.

Logistic regression RF SVM XGB
Model performance
AUROC (95% CI) 0.669 (0.536–0.803) 0.742 (0.637–0.835) 0.675 (0.561–0.789) 0.745 (0.631–0.847)
AUPRC (95% CI) 0.182 (0.104–0.274) 0.224 (0.119–0.397) 0.182 (0.102–0.337) 0.240 (0.125–0.424)
F1 score (95% CI) 0.853 (0.783–0.881) 0.888 (0.845–0.925) 0.865 (0.821–0.905) 0.896 (0.857–0.931)
Accuracy (95% CI) 0.753 (0.677–0.819) 0.805 (0.734–0.865) 0.773 (0.698–0.836) 0.818 (0.748–0.876)
Sensitivity (95% CI) 0.815 (0.739–0.876) 0.881 (0.815–0.931) 0.830 (0.755–0.889) 0.889 (0.823–0.936)
Specificity (95% CI) 0.316 (0.126–0.566) 0.263 (0.091–0.512) 0.368 (0.163–0.616) 0.316 (0.126–0.566)
PPV (95% CI) 0.194 (0.074–0.375) 0.238 (0.082–0.472) 0.233 (0.099–0.423) 0.286 (0.113–0.522)
NPV (95% CI) 0.894 (0.824–0.943) 0.895 (0.830–0.941) 0.903 (0.837–0.949) 0.902 (0.839–0.947)
Brier score
Overall 0.181 0.138 0.158 0.141
Reliability 0.086 0.038 0.058 0.054
Resolution 0.013 0.008 0.008 0.021
Uncertainty 0.108 0.108 0.108 0.108

AHREs atrial high-rate episodes, AUPRC area under the precision-recall curve, AUROC area under receiver operating characteristic, CI confidence interval, NPV negative predictive value, PPV positive predictive value, RF random forest, SVM support vector machine, XGB extreme gradient boosting.