A comparison of random forest and logistic regression shows why model choice depends on the data, prediction task and users’ need for interpretability. Results from an enrollment-yield study illustrate how similar overall accuracy can conceal important differences in classification.