Table 2.

Alternative splits of classification trees and improvement scores for each criterion.

	Best split		First alternative		Second alternative		Third alternative		Fourth alternative
	Criterion	I	Criteria	I	Criteria	I	Criteria	I	Criteria	I
Node 1	Wet matted < 0.75	13.72	Abdomen < 0.25	12.76	Thorax < 0.25	10.50	Hind leg < 0.25	9.05	Face color < 0.25	8.71
Node 2	Abdomen < 0.25	5.26	Camera category	4.49	Head color < 0.25	3.83	Thorax < 0.25	3.54	Hind leg < 0.25	3.41
Node 4	Camera category	4.45	Head color < 0.25	2.77	Hind leg < 0.75	1.62	Thorax < 0.25	1.58	Abdomen < 0.75	1.47
Node 5	n color morphs < 15.5	1.43	Face color < 0.25	1.33	Hind leg < 0.25	0.87	Head color < 0.25	0.85	Striped bee < 0.25	0.81
Node 9	Thorax side < 0.25	3.46	Face color < 0.25	1.34	Cuckoo	1.30	n color morphs < 8.5	1.05	Thorax < 0.75	0.76

I = Improvement Score, which is a relative measure of how much each split improves the homogeneity of the sub nodes. For instance, at node 1, splitting based on “wet and matted” improved the homogeneity of the resulting sub nodes by an index of 13.72 whereas splitting by “abdomen” improved the homogeneity by slightly less (12.76). Photo quality characteristics including “wet matted” and the visibility of key features like the abdomen were ocularly scored between 0 and 1 (clearly visible).