BMC Medical Informatics and Decision Making

Table 3 Comparison of CV-AUCs of generalized linear models (GLM) and random forest (RF), with and without clinical covariates

From: Improving random forest predictions in small datasets from two-phase sampling designs

	GLM		RF
	No covariates	Covariates	No covariates	Covariates
All markers	0.810	0.813	0.808	0.824
T cell markers	0.781	0.793	0.806	0.819
Antibody markers	0.759	0.768	0.729	0.711
No markers	0.500^a	0.624	0.500*	0.443

Screening is applied for both GLM and RF, but class balancing is not done for RF. Inverse sampling probability weights are used in both GLM and RF training
^aDenotes theoretical values

Back to article page

ISSN: 1472-6947

Contact us

General enquiries: journalsubmissions@springernature.com