Table 4.

Classification performance of models with various implementations

Category	Implementation	Accuracy	Precision	F1	Cohen’s kappa
Graph based (baseline)	LP	0.734 ± 0.0175	0.763 ± 0.0479	0.674 ± 0.0269	0.642 ± 0.0241
	Randomized LP (20)	0.652 ± 0.0465	0.709 ± 0.0277	0.584 ± 0.0453	0.528 ± 0.0741
	Randomized LP (90)	0.686 ± 0.0318	0.702 ± 0.0519	0.624 ± 0.0394	0.577 ± 0.0477
Model 2 (supervised)	Network depth 16	0.679 ± 0.0375	0.67 ± 0.0623	0.652 ± 0.046	0.6 ± 0.0491
	Network depth 32	0.683 ± 0.0368	0.671 ± 0.0661	0.656 ± 0.0458	0.6 ± 0.0486
	Network depth 64	0.667 ± 0.0431	0.628 ± 0.0784	0.626 ± 0.0554	0.572 ± 0.0601
Model 3 (supervised)	Network depth 16	0.787 ± 0.0369	0.804 ± 0.0376	0.774 ± 0.0354	0.733 ± 0.0439
	Network depth 32	0.808 ± 0.045	0.819 ± 0.034	0.798 ± 0.0448	0.761 ± 0.052
	Network depth 64	0.816 ± 0.0384	0.822 ± 0.0446	0.807 ± 0.0425	0.768 ± 0.0498
Model 4 (semisupervised)	Network depth 16	0.784 ± 0.046	0.789 ± 0.0419	0.776 ± 0.0458	0.731 ± 0.0542
	Network depth 32a	0.824 ± 0.0329	0.832 ± 0.0302	0.816 ± 0.0342	0.781 ± 0.04
	Network depth 64	0.815 ± 0.0252	0.814 ± 0.0265	0.804 ± 0.0262	0.768 ± 0.0306

LP = label spreading.

^aBest performing model.