Table 3. Mean Change in RMSEP for Each Prediction Model by Different Preprocessing Methods^a.

		mean change in RMSEP
preprocessing technique		PLSR	SVR	MLP	CNN	gradient boost
normalization	Mie baseline removal	–0.3326	–0.1252	–0.0063	–0.0087	–0.0071
	min-max scaling	–0.0866	0.3918	0.0088	0.0144	0.0130
	Z-score	0.0732	1.0256	0.0282	0.0273	0.0240
dimensionality reduction	FA	–0.0047	–0.3354	–0.0109	0.0249	0.0055
	ICA	–0.0037	–0.3356	0.0312	0.0443	0.0114
	isomap	0.0229	0.6478	0.0108	0.0316	0.0239
	LLE	–0.0021	–0.4042	0.0305	0.0080	0.0097
	NMF	0.0001	–0.4179	0.0118	0.0129	0.0080
	PCA	–0.0026	0.9325	–0.0050	0.0133	0.0119

^aRMSEP values are reported as dimensionless mass ratios.