Table 2.

Comparison between our model and prior studies.

Dataset	References	Size of Dataset	SE	SP	ACC	BACC
Training	Present study	1254 (636+/618−)	0.818	0.748	0.783	0.783
	[37]	1241 (683+/558−)	0.799	0.603	0.711	0.701
	[30]	978 (571+/407−)	0.948	0.585	0.797	0.767
	[33]	996 (541+/455−)	0.680	0.610	0.650	0.645
Test	[37]	83 (66+/17−)	0.879	0.647	0.831	0.763
			(0.909)	(0.529)	(0.831)	(0.719)
	[30]	85 (58+/27−)	0.707	0.815	0.741	0.761
			(0.848)	(0.345)	(0.682)	(0.597)
	[33]	67 (28+/39−)	0.786	0.590	0.672	0.688
			(0.536)	(0.641)	(0.597)	(0.588)
	Present study (Entire external test set)	204 (125+/79−)	0.773	0.658	0.730	0.716
	Present study (Reverse validation)	312 (0+/312−)	-	-	0.689	-
					(0.301)

In column of size of dataset, “+” and “−” denote the number of DILI-positives and DILI-negatives, respectively. For test, indicators within and outside parentheses were provided by prior studies and our model, respectively. We also investigated the performance of our model against the entire external test set which consisted of the external test sets provided by Ai et al., Zhang et al., and Kotsampasakou et al.