Table 2. Parameters optimization search space for the GCNN model.

Parameter	Search space (phase 1)
Graph convolution layers	[64, 64], [128, 64, 32], [64, 64, 64], or [256, 128, 64, 32]
Dense layer size	64, 128, or 256
Batch size	16, 32, or 64
Dropout	X ∼ U(0, 0.5)
Learning rate	log(X) ∼ U(−8.0, −3.0)
Weight decay	10−5, 10−4, or 10−3
Activation function	Hyperbolic tangent (tan h) or rectified linear unit (ReLU)