An image database of Drosophila melanogaster wings for phenomic and biometric analysis

GigaScience

Table 5 Classification accuracy of machine learning algorithms compared with BioCAT for predicting sex

Method	Training images	Testing images	Sex (± Standard error)
BioCAT	Olympus 40×	Olympus 40×	85.0 %
	Olympus 40×	Olympus 20×	50.0 %
	Olympus 40× cropped	Olympus 20× cropped	50.0 %
	Leica 40× cropped	Leica 40× cropped	93.0 %
	Olympus 40× cropped	Leica 40× cropped	73.7 %
	Olympus & Leica 40× cropped	Olympus 40× cropped	73.3 %
	Olympus & Leica 40× cropped	Leica 40× cropped	86.0 %
Landmarks	Olympus 40× landmarks	Olympus 40× landmarks	98.2 % (±1.6)
	Olympus 40× landmarks	Olympus 20× landmarks	81.2 % (±1.4)
	Leica 40× landmarks	Leica 40× landmarks	97.8 % (±0.69)
	Olympus 40× landmarks	Leica 40× landmarks	79.1 % (±1.3)

Machine learning algorithms using landmark and semi-landmark features, compared with Hessian features extracted by BioCAT, trained and tested across microscopes and magnifications

ISSN: 2047-217X