Figure 1. AUC-based performance following removal or inclusion of individual four-feature types in the optimal feature set.

The graphs illustrate the impact on prediction performance of the Crysalis first-level models for all five prediction classes. (A) Performance of the trained models following removal of the corresponding feature type. (B) Performance of the trained models using the individual feature type. All predicted models are compared with the best models that were trained using the optimal feature sets.