Rapid Characterization of Tanshinone Extract Powder by Near Infrared Spectroscopy

Table 9

The best preprocessing methods for near infrared spectra of the least squares support vector machine model for different quality attributes.

Index	Processing method	gam	sig^2	Calibration set		Validation set
					BIAScal		RMSEP	RPD	BIASpre
Cc/mg·g^−1	S-G 1st	2299.6	50195	0.9980	9.3 × 10^−4	0.9985	0.0020	15	6.5 × 10^−4
IIA/mg·g^−1	S-G smooth	15042	5963.3	0.9996	6.9 × 10^−4	0.9978	0.0010	12	7.2 × 10^−4
SSA/m^2·g^−1	Normalization	17.4749	2541.4	0.9207	0.023	0.9661	0.017	2.5	0.016
/m	S-G 1st	212.18	4476.9	0.9908	0.42	0.9723	0.67	3.2	0.49
/m	Raw	25.830	2222.9	0.9604	3.9	0.9795	3.1	4.3	2.69
/m	Normalization	1928.5	2106.5	0.9835	3.8	0.9276	7.8	2.7	5.8
/g·cm^−3	S-g smooth	6.9355	111.63	0.9851	0.011	0.8875	0.020	2.2	0.018
Moisture/%	Baseline	2042.3	1970.5	0.8900	0.28	0.9336	0.30	2.9	0.25

Gam and sig² are two tuned hyperparameters of LS-SVM model. and represent correlation coefficient for calibration and validation sets, respectively. RMSEP represents the root mean square error of prediction. BIAS_cal and BIAS_pre represent bias for calibration and validation, respectively. RPD means relative predictive deviation.
Cc and IIA mean the content of cryptotanshinone and tanshinone IIA, respectively. SSA, mean specific surface area and tapped density, respectively. The , and values represent the maximal particle size diameters that include 10%, 50% and 90% of the particles, respectively.