Figure 7.

Variable of importance by randomForest for classification of each stage and for overall accuracy. Leftmost four panels list the variables that are important for classifying each stage, while the right panel lists the variables important for an overall accuracy of classification of all stages. For both randomForest and Caret packages, endometriosis is the most important variable among Stage 1 participants, and the overall accuracy also specifies endometriosis, mass, breathing + ascites, and menstruation problem as the most important ones among all. The number of trees evaluated was from 500 to 10,000, and 5,000 trees was the most stable in terms of OOB estimate of error rate. Caret package has the randomForest imbedded in.