Figure 4. Effect of negative data selection on model performance.

Accuracy distribution at 100 fold MCCV for the eight positive datasets. All other variables were summarized (e.g., six learners).