A Progressive Combined Variable Selection Method for Near-Infrared Spectral Analysis Based on Three-Step Hybrid Strategy

Table 2

Results of synergy interval partial least squares selection subintervals.

Number of intervals	Number of combinations	Interval	nLV	Corresponding band (nm)	RMSE (%)
10	2	5, 9	5	1,660–1,798, 2,220–2,358	2.165
	3	4, 5, 9	4	1,520–1,798, 2,220–2,358	3.714
	4	4, 5, 8, 9	4	1,520–1,798, 2,080–2,358	4.219
20	2	9, 18	6	1,660–1,728, 2,290–2,358	3.747
	3	9, 17, 18	8	1,660–1,728, 2,220–2,358	2.360
	4	9, 10, 17, 18	5	1,660–1,798, 2,220–2,358	2.165
30	2	12, 26	7	1,626–1,670, 2,270–2,314	4.075
	3	13, 14, 26	9	1,672–1,762, 2,270–2,314	2.525
	4	13, 14, 25, 26	6	1,672–1,762, 2,224–2,314	1.806

Notes. nLV – number of latent variables. “nm” – nanometer. RMSE – root mean square error. The synergy interval partial least squares method divided the whole spectrum into 10, 20, and 30 intervals. Based on three division methods, 2, 3, and 4 subintervals were combined, respectively. The RMSE value in each case was obtained. The spectral bands corresponding to the subinterval were the selected variables.