Early Recognition of Skin Malignancy in Images Based on Convolutional Networks by Using Dynamic System Model

Table 6

Table of experiments performed with VGG-16, applying data augmentation and different subdivisions of the dataset. Best results are highlighted in bold.

Experiment	Basic network	Data augmentation			Proportion	Best accuracy
		Flip H	Flip V	Jitter	Train/val ratio (%)	Train	Val
1	VGG-16	No	No	No	80-20	7.36E-01	6.49E-01
2					70-30	7.35E-01	6.74E-01
3					55-45	7.27E-01	6.93E-01
4		Yes			80-20	7.34E-01	6.55E-01
5					70-30	7.33E-01	6.76E-01
6					55-45	7.26E-01	6.94E-01
7			Yes		80-20	7.34E-01	6.53E-01
8					70-30	7.34E-01	6.73E-01
9					55-45	7.27E-01	6.94E-01
10				Yes	80-20	7.17E-01	6.37E-01
11					70-30	7.19E-01	6.61E-01
12					55-45	7.14E-01	6.90E-01