Table 3 Correlation between test area-under-the-curve and sample size above and below auto-encoder loss inflection points for with varying levels of label accuracy.
From: Autoencoders for sample size estimation for fully connected neural network classifiers
Pre-MCSE | Post-MCSE | |||||
|---|---|---|---|---|---|---|
% Mislabled | R2 | Kendall’s τ | Spearman’s ρ | R2 | Kendall’s τ | Spearman’s ρ |
0.10 | 0.006 | 0.074 | 0.092 | 0.686 | 0.703 | 0.842 |
1.00 | 0.003 | –0.025 | –0.032 | 0.710 | 0.721 | 0.865 |
2.00 | 0.021 | 0.100 | 0.126 | 0.706 | 0.694 | 0.839 |
5.00 | 0.002 | 0.054 | 0.070 | 0.687 | 0.654 | 0.812 |
10.0 | 0.001 | 0.002 | −0.001 | 0.714 | 0.681 | 0.827 |
20.0 | 0.012 | 0.084 | 0.106 | 0.644 | 0.619 | 0.768 |
30.0 | 0.004 | 0.028 | 0.038 | 0.465 | 0.498 | 0.636 |
40.0 | 0.000 | −0.017 | −0.022 | 0.389 | 0.418 | 0.557 |
50.0 | 0.024 | −0.105 | −0.137 | 0.385 | 0.354 | 0.471 |
