Table 5.

Comparison of gradient boosting prediction accuracy using limited numbers of features for different feature selection methods. Bold entries denote highest accuracy for a fixed maximum number of features.

Feature Selector	≤ 5 Features	≤ 10 Features	≤ 20 Features
LASSO stability selection	0.653 ± 0.0322	0.653 ± 0.0322	0.687 ± 0.017
MDI using random forest	0.680 ± 0.030	0.715 ± 0.037	0.715 ± 0.037
MDI using gradient boosting	0.666 ± 0.016	0.666 ± 0.016	0.673 ± 0.018
Maximum depth limitation	0.625 ± 0.004	0.639 ± 0.023	0.687 ± 0.025