Rapid measurement of soluble xylo-oligomers using near-infrared spectroscopy (NIRS) and multivariate statistics: calibration model development and practical approaches to model optimization

Table 3 Summary of model performance results by process stream location and spectral range used in modeling

Performance parameter		XOS (g/L)				Total solids (g/L)
		Early/waste stream sampling locations		Late stream sampling locations		Early/waste stream sampling locations		Late stream sampling locations
		1350–2500 nm	2100–2450 nm	1350–2500 nm	2100–2450 nm	1350–2500 nm	2100–2450 nm	1350–2500 nm	2100–2450 nm
Training	R²	0.97	0.96	1.00	1.00	0.99	0.99	1.00	1.00
Training	RMSEC	0.24	0.28	4.17	3.76	0.30	0.22*	3.72	4.52
Cross validation	R²	0.46	0.71	0.99	1.00	0.46	0.76*	1.00	1.00
Cross validation	RMSECV	1.16	0.78*	4.70	4.08	2.26	1.37*	4.08	4.88
Independent validation	R²	0.52	0.60	0.99	0.99	0.69	0.66	0.99	0.99
Independent validation	RMSEP	0.67	0.63	4.81	4.98	0.97	1.07	5.02	5.86

Results are shown for prediction of XOS concentration (g/L) and total solids concentration (g/L). Reducing the spectral range used in the model to 2100-2450nm does not substantially affect the quality of either the XOS concentration or the total solids concentration models. Asterisks (*) indicate statistically significant differences in truncated-range model compared to corresponding full-range model

ISSN: 2731-3654