. Author manuscript; available in PMC: 2021 Jul 27.

Published in final edited form as: Proc AAAI Conf Artif Intell. 2020 Apr 3;34(5):8180–8187. doi: 10.1609/aaai.v34i05.6331

Table 2:

Performance comparisons using different configurations in the multi-filter and residual convolutional layers. k denotes the kernel sizes k₁, k₂, …, k_m and p denotes the residual block number.

Model	Config	MIMIC-III, full codes			MIMIC-III, top-50 codes
Model	Config	P@8	Micro-F1	Macro-F1	P@5	Micro-F1	Macro-F1
CNN	k=9	0.706	0.508	0.053	0.590	0.592	0.519

	k=5,9,15	0.731	0.534	0.061	0.616	0.633	0.556
k=3,5,9,15,19	0.735	0.542	0.067	0.630	0.646	0.576
k=3,5,9,15,19,25	0.736	0.545	0.068	0.633	0.652	0.584

ResCNN	p=1	0.714	0.532	0.063	0.618	0.645	0.560
	p=2	0.713	0.532	0.059	0.589	0.601	0.531
	p=3	0.710	0.529	0.059	0.575	0.585	0.500

MultiResCNN	k=3,5,9,15,19,25	0.741	0.561	0.073	0.638	0.673	0.608
MultiResCNN	p=1	0.741	0.561	0.073	0.638	0.673	0.608