Table 2.

Results of the discriminative model in distinguishing GBM from PCNSL in the training and validation group.

Classifier	Selection method	Training group				Validation group
		AUC	Accuracy	Sensitivity	Specificity	AUC	Accuracy	Sensitivity	Specificity
LDA	Distance correlation	0.992	0.993	0.996	0.990	0.978	0.979	0.982	0.976
	RF	0.970	0.968	0.935	0.990	0.964	0.957	0.906	0.990
	LASSO	0.997	0.996	0.992	0.995	0.977	0.971	0.955	0.989
	Xgboost	0.791	0.810	0.995	0.740	0.750	0.789	0.995	0.735
	GBDT	0.972	0.970	0.939	0.996	0.956	0.950	0.892	0.995
SVM	Distance correlation	0.957	0.962	0.998	0.934	0.959	0.964	0.997	0.943
	RF (over-fitting)	1	1	1	1	0.5	0.585	1	0.943
	LASSO	0.843	0.835	0.747	0.966	0.822	0.789	0.671	0.965
	Xgboost (over-fitting)	0.5	0.541	0.747	0.967	0.5	0.586	0.671	0.965
	GBDT (over-fitting)	1	1	1	1	0.5	0.586	0.670	0.965
LR	Distance correlation	0.977	0.956	0.961	0.949	0.933	0.927	0.941	0.911
	RF (over-fitting)	1	0.547	1	0.592	0.511	0.515	0.551	0.596
	LASSO	0.959	0.988	0.942	0.981	0.975	0.966	0.975	0.964
	Xgboost (over-fitting)	0.959	0.988	0.942	0.981	0.5	0.5	0.542	0.586
	GBDT (over-fitting)	0.951	0.562	0.954	0.592	0.538	0.515	0.577	0.596

AUC, area under curve; RF, random forest; LASSO, least absolute shrinkage and selection operator; Xgboost, eXtreme gradient boosting; GBDT, Gradient Boosting Decision Tree; LDA, linear discriminant analysis; SVM, support vector machine; LR, logistic regression.