Table 1 Results.
From: Bias in error estimation when using cross-validation for model selection
Classifier type | Percentage of times training error<30% | Percentage of times bias>20% | Mean CV error estimate for optimized classifier | Mean true error of optimized classifier | Mean bias |
|---|---|---|---|---|---|
Shrunken centroids on "null" data | 18.5% | 22.2% | 37.8% | 50.0% | -12.2% |
SVM on "null" data | 22.2% | 25.3% | 41.7% | 50.0% | -8.3% |