Multiscale Latent Variable Regression

Table 2

Comparison between the output prediction mean square errors obtained by MSLVR at multiple scales (the numbers in parenthesis indicate the percentages that each scale is selected as optimum using the CVMSE criterion).

Scale	PCR MSE(×10^7) SNR=10	PLS MSE(×10^7) SNR=10	PCR MSE(×10^8) SNR=50	PLS MSE(×10^8) SNR=50
𝑗=0	0.907 (0)	0.822 (0)	0.533 (0)	0.487 (0)
𝑗=1	0.705 (0)	0.643 (0)	0.484 (0)	0.436 (0)
𝑗=2	0.595 (0)	0.537 (0)	0.449 (0)	0.404 (0)
𝑗=3	0.523 (3)	0.463 (1)	0.416 (11)	0.373 (13)
𝑗=4	0.479 (97)	0.422 (99)	0.401 (89)	0.364 (87)
𝑗=5	0.636 (0)	0.584 (0)	0.572 (0)	0.541 (0)