Table 3. Predictive performance of seven machine learning methods on four FT NIR spectral data resulting from 7-fold cross-validation.

FT NIR spectral data	Machine learning models	AUC	Misclassification	R2
Raw data	XGBoost	0.964	0.089	0.661
	Bootstrap Forest	0.940	0.123	0.474
	Boosted Tree	0.937	0.133	0.533
	Neural Networks	0.904	0.191	0.422
	Decision Tree	0.821	0.227	0.250
	Support Vector Machine	0.735	0.338	0.108
	PLS-DA	0.766	0.131	0.406
	Mean (7 models)	0.867	0.176	0.408
	Mean (top 4 models)	0.936	0.134	0.523
SG (2nd polynomial) transformed data	XGBoost	0.965	0.081	0.676
	Bootstrap Forest	0.950	0.115	0.475
	Boosted Tree	0.940	0.123	0.581
	Neural Networks	0.914	0.167	0.435
	Decision Tree	0.850	0.193	0.328
	Support Vector Machine	0.734	0.338	0.106
	PLS-DA	0.749	0.249	0.371
	Mean (7 models)	0.872	0.181	0.424
	Mean (top 4 models)	0.942	0.122	0.542
SG (2nd polynomial) + 1st derivative transformed data	XGBoost	0.956	0.086	0.610
	Bootstrap Forest	0.946	0.123	0.488
	Boosted Tree	0.936	0.124	0.534
	Neural Networks	0.891	0.190	0.395
	Decision Tree	0.843	0.247	0.271
	Support Vector Machine	0.795	0.253	0.227
	PLS-DA	0.762	0.236	0.369
	Mean (7 models)	0.876	0.180	0.413
	Mean (top 4 models)	0.932	0.131	0.507
SG (2nd polynomial) + 2nd derivative transformed data	XGBoost	0.900	0.173	0.350
	Bootstrap Forest	0.890	0.185	0.353
	Boosted Tree	0.880	0.184	0.376
	Neural Networks	0.844	0.241	0.277
	Decision Tree	0.821	0.259	0.122
	Support Vector Machine	0.733	0.287	0.174
	PLS-DA	0.770	0.227	0.352
	Mean (7 models)	0.834	0.222	0.286
	Mean (top 4 models)	0.879	0.196	0.339