Model selection based on logistic regression in a highly correlated candidate gene region

BMC Proceedings

Table 2 Evaluation of model selection

		Average Prediction Error (SE)^a
Selection methods	No. selected SNPs	Without regularization	Ridge penalty	Random forests
Stepwise/AIC	9	0.1536 (0.0049)	0.1506 (0.0044)	0.1586 (0.0068)
Lasso/AIC	29	0.1572 (0.0051)	0.1450 (0.0046)	0.1469 (0.0059)
Lasso/CV	17	0.1461 (0.0052)	0.1428 (0.0060)	0.1558 (0.0052)
Bayesian	2	0.1306 (0.0052) ^b	0.1306 (0.0052)	0.1336 (0.0043)
Trait locus DR/C*			0.1572 (0.0058)

^aAverage prediction error was calculated from Replicates 2 to 10 using the optimally selected set of SNPs from Replicate 1.
^bBold indicates the minimum average prediction error.

ISSN: 1753-6561