Table 4.

Prediction performance benchmarking for the prediction task of “Performance against Physical Attributes" on the test set.

Property	Data size	Base	MAE of best SC model	MAE of best TL model
KLU	28,056	18.77	10.79	11.15
KAA	28,171	5.234	2.722	2.832
BgOptb	28,163	0.988	0.279	0.251
Deltae	28,155	0.850	0.135	0.125
Encut	28,108	246.25	76.99	82.26
Ehull	27,297	0.131	0.058	0.050
Magoszi	25,844	1.225	0.438	0.405
Magout	25,357	1.176	0.393	0.376
Eps	25,150	3.829	1.280	1.262
PPF	16,250	650.5	495.0	494.0
NPF	16,250	658.1	484.2	512.3
Pem300k	16,763	1.918	1.086	1.228
Nem300k	16,760	1.918	1.086	1.197
PSB	14,439	163.30	56.49	60.25
NSB	14,144	108.69	48.93	54.53
Meps	11,349	4.905	1.784	1.776
MaxM	10,963	285.32	57.38	65.69
MinM	10,930	40.89	24.12	23.51
ETC11	10,839	81.66	34.53	33.19
ETC12	10,759	44.96	16.60	17.23
ETC13	10,846	42.54	14.09	13.93
ETC22	10,832	84.06	34.62	31.35
ETC33	10,856	84.12	35.82	34.22
ETC44	9986	29.55	15.23	14.63
ETC55	9755	26.61	12.40	11.74
ETC66	9739	27.59	13.53	13.10
BulkKV	10,743	49.11	11.83	11.28
ShearGV	10,209	24.28	11.26	11.01
BgMbj	7296	1.911	0.555	0.534
Spillage	3866	0.501	0.410	0.373
SLME	3006	9.439	7.193	6.420
MaxIrM	2302	426.0	87.67	103.4
MinIrM	2268	66.16	38.52	44.45
PMDiEl	2126	5.757	3.911	2.715
PMDi	2126	6.977	4.336	3.561
PMDiIo	2126	2.577	0.847	0.774
PMEij	1123	0.520	0.436	0.367
PMDij	689	46.47	26.46	23.46
Exfoli	557	62.93	54.26	51.56

The table shows the test MAE of the best model selected using Supplementary Table 9 (based on validation MAE) when run on the test set for each of the target materials properties. The selected modeling configurations are listed in Supplementary Table 10. The SC models are allowed to use PA as input as well, while TL models only use EF-based inputs. The lowest MAE values in each row are highlighted in bold.