Figure 1. Feature optimization for GBM. Plotted are the areas under the curves (AUCs) of the receiver operator characteristics acquired through our incremental feature selection process.

Three sets of AUCs are computed from the 10-fold cross-validation (CV) of the training set GBM.Ex (dotted line) and the independent validation (IV) of 2 test sets, GBM.S1 and GBM.S2 (solid and dashed line). On the x-axis are features that are incrementally selected. The dashed box marks the peaks of the cross-validation AUC, which corresponds to the optimal feature set used for CanDrA.