. Author manuscript; available in PMC: 2024 Feb 9.

Published in final edited form as: Sci Stat Database Manag. 2023 Aug 27;2023:1. doi: 10.1145/3603719.3603737

Table 3:

Model performance comparison. Average results for AUC, Precision (Pr), Recall (Rc), and F1-Score (F1) were computed using 5 times 5-Fold cross-validation for the two models. Performance for each class using micro-averaging (ALL).

Model	Setting	Class	AUC	Pr	Rc	F1
Three-Class MLP	Full set	GBM	.881	.830	.806	.818
		LYM	.859	.681	.705	.693
		MET	.920	.795	.803	.799
		ALL	.911	.790	.789	.789
	AE 5 latent	GBM	.894	.737	.778	.757
		LYM	.846	.667	.500	.571
		MET	.925	.800	.833	.816
		ALL	.887	.756	.760	.756
	AE 15 latent	GBM	.910	.801	.813	.807
		LYM	.897	.749	.640	.690
		MET	.923	.781	.810	.795
		ALL	.909	.783	.784	.783
	AE 50 latent	GBM	.910	.839	.806	.822
		LYM	.868	.639	.575	.605
		MET	.889	.776	.825	.800
		ALL	.899	.777	.779	.777
Two-Stage	Stage 1	GBM	.852	.828	.871	.849
	Stage 1	REST	.852	.750	.682	.714
	Stage 2	LYM	.864	.667	.581	.621
	Stage 2	MET	.864	.880	.913	.896
	Combined	ALL	-	.717	.712	.713