Table 3. Performance of the GCN and LM-GCN Models on Various Data Sets^a.

		GCN model				LM-GCN model
Data set		Train		Validation		Train		Validation
Train	Eval.	↑r2	↓loss	↑r2	↓loss	↑r2	↓loss	↑r2	↓loss
D1	D1	0.819	0.182	0.813	0.188	0.808	0.158	0.805	0.194
D2	D2	0.574	0.427	0.552	0.435	0.667	0.310	0.651	0.356
D3	D3	0.502	0.498	0.497	0.489	0.642	0.347	0.625	0.368
D4	D4	0.596	0.405	0.588	0.405	0.601	0.399	0.597	0.405
D5	D5	0.786	0.213	0.778	0.227	0.852	0.152	0.855	0.147
D5	D1	–	–	0.602	0.403	–	–	0.638	0.361
D4	D3	–	–	0.396	0.587	–	–	0.449	0.542

^aThe architecture of the LM-GCN model is shown in Figure 6. The size of the data sets is 10,000 examples, with 80:20 split for training and validation, respectively.