Skip to main content

View full-text article in PMC

. 2023 Feb 7;25:e43734. doi: 10.2196/43734

Table 2.

Model performance to predict amiodarone-induced thyroid dysfunction.

Models		Major indices					Minor indices
		AUPRC^a	Recall	F₁-score	G-mean^b	Accuracy		Precision	AUROC^c
XGBoost^d
	Raw	0.742	0.607	0.670	0.769	0.932		0.748	0.936
	B-SMT^e	0.751	0.756	0.688	0.845	0.923		0.632	0.934
	ENN^f	0.741	0.702	0.661	0.815	0.918		0.624	0.939
	B-SMT ENN^g	0.730	0.673	0.641	0.797	0.914		0.611	0.924
AdaBoost^h
	Raw	0.643	0.516	0.593	0.708	0.919		0.696	0.923
	B-SMT	0.654	0.858	0.589	0.861	0.864		0.448	0.921
	ENN	0.635	0.742	0.640	0.829	0.905		0.562	0.922
	B-SMT ENN	0.624	0.822	0.583	0.847	0.867		0.452	0.914
K-nearest neighbor
	Raw	0.500	0.149	0.249	0.385	0.898		0.759	0.835
	B-SMT	0.470	0.829	0.330	0.699	0.617		0.206	0.816
	ENN	0.393	0.255	0.337	0.496	0.886		0.496	0.825
	B-SMT ENN	0.467	0.822	0.338	0.709	0.635		0.213	0.813
Logistic regression
	Raw	0.294	0.073	0.118	0.26	0.877		0.313	0.798
	B-SMT	0.303	0.804	0.377	0.742	0.698		0.246	0.806
	ENN	0.300	0.229	0.260	0.462	0.852		0.300	0.811
	B-SMT ENN	0.305	0.778	0.389	0.746	0.722		0.259	0.803

^aAUPRC: area under the precision-recall curve.

^bG-mean: geometric mean.

^cAUROC: area under the receiver operating characteristic curve.

^dXGBoost: extreme gradient boosting.

^eB-SMT: borderline synthesized minority oversampling technique.

^fENN: edited nearest neighbors.

^gB-SMT ENN: hybrid oversampling with borderline synthetic minority oversampling technique and undersampling with edited nearest neighbor.

^hAdaBoost: adaptive boosting.