Table 4. Mean squared error values from trials with linear, LASSO (with and without CV), ridge (with and without CV), and stepwise regressions, and from trials of the neural network model are shown for data sets 3A/3B and 4A/4B and 5A/5B. Note that some values are from identical regressions, such as the linear regression of 5A being identical to the stepwise regression of 3A.

	5A	4A	3A	5B	4B	3B
Linear	0.0026 a	0.0026	0.0026	0.0061 a	0.013	0.0059
Ridge (α = 0.01)	—	0.0026	0.0026	—	0.013	0.0060
LASSO (α = 0.01)	—	0.0092 a	0.0092 a	—	0.020 a	0.020 a
NN (with CV)	0.0056	0.0038	0.0081 b	0.0034	0.013	0.0034 b
Ridge CV	—	0.0033	0.0028	—	0.014	0.0088
LASSO CV	—	0.0032	0.0026	—	0.016	0.0079
Stepwise	—	—	0.0026 a	—	—	0.0061 a
Gaussian process	—	0.013	0.0079	—	0.060	0.0082

^aThese models have identical models with the same MSE's.

^bThe NN MSE's for 3A and 3B shown here are calculated differently than those for Table 3, as explained in the Methods.