Fig. 4: Variable Importance (VIMP) for the best-fitted random forest (left) is presented for rhinitis.

Blue bars indicate the importance of variables (positive VIMP) relative to their lengths, red bars indicate presumably noise variables (negative VIMP). The best combination for the hyper-parameters was found to be nodesize = 10, nsplit = 2, and mtry = 3 with the lowest OBB error estimate of 2%. Minimal depth (right) refers to the topology of the forest. Low minimal depth indicates important variables. The dashed line represents the threshold of the maximum value for variable selection.