Table 3.

Prediction performance of BSI using different algorithms.

Dataset	Algorithms 1	AUROC (95% CI)	Sensitivity (95% CI)	Specificity (95% CI)	Brier Score
Validation dataset	LR	0.709 (0.679–0.737)	0.679 (0.624–0.728)	0.660 (0.625–0.695)	0.218
	SVM	0.728 (0.699–0.756)	0.578 (0.522–0.632)	0.779 (0.747–0.809)	0.195
	MLP	0.735 (0.707–0.761)	0.494 (0.438–0.549)	0.832 (0.803–0.858)	0.231
	XGBoost	0.825 (0.802–0.849)	0.724 (0.672–0.771)	0.777 (0.744–0.806)	0.165
	RF	0.855 (0.832–0.877)	0.565 (0.509–0.619)	0.927 (0.905–0.944)	0.139
Testing dataset	LR	0.685 (0.653–0.715)	0.615 (0.558–0.670)	0.644 (0.609–0.679)	0.223
	SVM	0.704 (0.673–0.733)	0.566 (0.508–0.623)	0.756 (0.723–0.786)	0.199
	MLP	0.668 (0.633–0.698)	0.406 (0.350–0.463)	0.811 (0.781–0.838)	0.254
	XGBoost	0.821 (0.795–0.843)	0.706 (0.651–0.756)	0.775 (0.743–0.804)	0.163
	RF	0.851 (0.824–0.872)	0.577 (0.519–0.633)	0.940 (0.921–0.955)	0.134

¹ LR: logistic regression, SVM: support vector machine, MLP: multi-layer perceptron, XGBoost: eXtreme Gradient Boosting, RF: random forest.