Table 2.

Hyperparameter tuning experiments.

Model	λ	Dropout	Learning rate	Layers	Hidden units	Accuracy (%)	Precision (%)	Recall (%)
GCN (Baseline)	None	0.5 (2,4,5 layer)	0.005	5	32/32/64/64/128	83.98 ± 3.2	84.59 ± 3.1	87.78 ± 6.4
GIN+Infomax	0.05	0.5	0.005	5	64	84.61 ± 2.9	86.19 ± 3.3	86.81 ± 4.9
GIN	0.0	-	-	-	-	84.41 ± 2.8	85.39 ± 2.6	87.60 ± 7.5
-	0.01	-	-	-	-	84.08 ± 2.2	86.72 ± 4.4	85.31 ± 5.5
-	0.1	-	-	-	-	84.51 ± 2.1	86.85 ± 4.5	86.06 ± 5.5
-	-	0.0	-	-	-	83.99 ± 3.4	85.78 ± 4.4	86.26 ± 6.1
-	-	-	0.01	-	-	83.13 ± 3.4	85.89 ± 3.4	84.01 ± 5.2
-	-	-	0.001	-	-	81.54 ± 3.3	85.45 ± 3.4	81.37 ± 7.3
-	-	-	-	4	-	83.11 ± 3.2	84.62 ± 2.8	85.70 ± 4.2
-	-	-	-	-	32	83.13 ± 3.4	85.20 ± 4.3	85.14 ± 5.5

Bold value indicates the saliency with respect to the input (24).