Convolutional Neural Network-Based Discriminator for Outlier Detection

Table 5

The results of the 5-seed discriminator for MNIST with data augmentation.

Noise ratio	10 (%)	50 (%)	90 (%)
Recall	84.82	93.78	97.56
F1 measure	56.54	93.83	98.58
False negative	0.17	1.8	8.77
False positive	60.4	9.9	1.8