Potential of machine learning approaches for predicting mechanical properties of spruce wood in the transverse direction

Journal of Wood Science

Official Journal of the Japan Wood Research Society

Table 3 Accuracy of regular CNN and density-informed CNN in prediction of MOE and MOR of specimen

Learning algorithm	Splitting method	Group	MOE (MPa)		MOR (MPa)
Learning algorithm	Splitting method	Group	R²	RMSE	R²	RMSE
Regular CNN	Method #1	Train	0.929 (0.013)	56 (5)	0.855 (0.032)	0.50 (0.05)
	Method #1	Test	0.735 (0.048)	108 (14)	0.536 (0.076)	0.90 (0.11)
	Method #2	Train_A2,B1,B2	0.967	40	0.945	0.28
		Test_A1	− 0.468	167	− 3.50	1.74
		Train_A1,B1,B2	0.941	57	0.882	0.51
		Test_A2	0.125	113	-0.064	0.75
		Train_A1,A2,B2	0.914	62	0.862	0.50
		Test_B1	0.328	162	− 0.880	1.14
		Train_A1,A2,B1	0.911	47	0.841	0.42
		Test_B2	0.200	238	− 1.691	1.99
Density-informed CNN	Method #1	Train	0.961 (0.031)	39 (15)	0.953 (0.008)	0.29 (0.03)
	Method #1	Test	0.859 (0.071)	77 (23)	0.812 (0.022)	0.54 (0.04)
	Method #2	Train_A2,B1,B2	0.947	50	0.972	0.20
		Test_A1	0.728	72	− 0.249	0.92
		Train_A1,B1,B2	0.968	42	0.864	0.55
		Test_A2	0.646	72	0.255	0.63
		Train_A1,A2,B2	0.986	25	0.931	0.35
		Test_B1	0.771	94	0.572	0.55
		Train_A1,A2,B1	0.967	29	0.927	0.29
		Test_B2	0.651	155	0.423	0.92