Table 1.

Predictive ability of models, expressed as squared correlation coefficient (${\text{Q}}^2$) between acd_logd values in ChEMBL database and predicted logD values for 100,000 test set compounds

Epsilon	Cost
	0.001	0.01	0.1	1	10
${10}^{-1}$	0.509	0.509	0.510
${10}^{-2}$	0.820	0.821	0.821	0.821
${10}^{-3}$	0.918	0.943	0.949	0.952	0.952
${10}^{-4}$	0.923	0.958	0.970	0.973	0.973
${10}^{-5}$		0.958	0.971	0.973	0.972

Bolditalic values indicate models with the highest predictive ability