Fig. 1

Experimental Design. Classification-based model fitting began with a unique combination of n _tree , m _try , and sampsize parameters in conjunction with training data, illustrated by the gray boxes. Each learned random forest model was used to predict the class of the validation data. Subsequently, AUC scores were calculated using the true class labels and these values were randomly subsetted into training and validation groups using 2/3 and 1/3 of the samples, respectively. In the second model fitting step, we evaluated whether AUC could be predicted from parameter sets alone. A RF regression model was fit using the parameters n _tree , m _try , and sampsize as variables and AUC as the response, illustrated by the blue boxes. Default settings were selected to train the RF regression models and AUC scores were predicted for the validation data. We evaluated the results using Spearman's and Lin’s correlation and determined the relative importance of each variable