Table 1.

The prediction power of the radiomic features extracted from target nodule images with different learning algorithms and feature selection methods over the balanced dataset.

Learning Algorithm	Target Radiomic Prediction Performance (AUROC)
	Feature Selection
	None	CST	Corr	LASSO	RELIEF	MI	PCA	FFS
Adab	0.889 ± 0.016	0.863 ± 0.021	0.881 ± 0.031	0.864 ± 0.026	0.671 ± 0.019	0.531 ± 0.037	0.868 ± 0.015	0.911 ± 0.016
DT	0.723 ± 0.011	0.733 ± 0.027	0.703 ± 0.019	0.711 ± 0.028	0.642 ± 0.013	0.523 ± 0.024	0.712 ± 0.026	0.730 ± 0.032
RF	0.871 ± 0.008	0.849 ± 0.025	0.846 ± 0.028	0.856 ± 0.023	0.765 ± 0.018	0.517 ± 0.031	0.862 ± 0.026	0.891 ± 0.011
KNN	0.850 ± 0.016	0.846 ± 0.016	0.807 ± 0.036	0.833 ± 0.017	0.735 ± 0.021	0.671 ± 0.089	0.846 ± 0.017	0.870 ± 0.023
SVM	0.777 ± 0.011	0.774 ± 0.029	0.752 ± 0.025	0.775 ± 0.027	0.751 ± 0.020	0.522 ± 0.040	0.775 ± 0.028	0.802 ± 0.008
LDA	0.655 ± 0.045	0.680 ± 0.032	0.785 ± 0.017	0.75 ± 0.027	0.741 ± 0.031	0.735 ± 0.018	0.771 ± 0.028	0.796 ± 0.011
QDA	0.778 ± 0.172	0.696 ± 0.181	0.747 ± 0.016	0.738 ± 0.024	0.753 ± 0.031	0.840 ± 0.020	0.752 ± 0.026	0.865 ± 0.006
Naive	0.763 ± 0.006	0.759 ± 0.023	0.742 ± 0.030	0.731 ± 0.022	0.756 ± 0.034	0.583 ± 0.046	0.739 ± 0.024	0.808 ± 0.010

For each feature selection algorithm, the highest value is marked in bold.