Comparison of GCV and fivefold cross-validation. The relation between prediction mean square error (PMSE) and smoothing parameter α
is shown for the three iterations (a-c) in the OLIN procedure applied to slide 16 of the apo AI experiment. The fivefold cross-validation was conducted for five random splits of the data. Mean values and standard errors of PMSE estimates are represented as black circles and error bars. PMSE estimates by GCV are represented by red squares. Generally, these estimates lie within the error margin of PMSE produced by fivefold cross-validation. The GCV-optimized value of α
was 0.3 for the first (a), 0.4 for the second (b), and 1.0 for the third (c) iteration.