Figure 3

Workflow to map sexual dimorphism: (A) calculate a neighborhood of 10 points around each semilandmark. (B) Use part of the sample (70%) as a training set. (C) Perform generalized procrustes analysis (GPA) on the training set. (D) Apply the predictive model to the testing set. (E) Calculate accuracy for each set (B–D) and repeat the process 100 times, randomly defining the training and testing sets. (B-D) At the end of the iterative process, calculate the average accuracy value associated with each semilandmark. (E) Convert the average accuracy values into a 3D color map.