From: To supervise or to self-supervise: a machine learning based comparison on credit supervision
Efficiency metrics | Description | Formulae |
---|---|---|
Accuracy | Number of correct predictions to total number of samples ratio | \(\frac{True\, Positives\,+\,True\, Negatives}{Total \,number\, of \,samples}\) |
Precision | Proportion of correct positive predictions, in relation to the total of positive predictions | \(\frac{True \,Positives}{False\, Positives\,+\,True \,Positives}\) |
F1 Score | Harmonic mean between precision and recall | \(2*\frac{1}{\frac{1}{precision}+\frac{1}{recall}}\) |
True positive rate (recall or sensitivity) | Proportion of correct positive predictions, in relation to all relevant samples, i.e., all positive samples | \(\frac{True \,Positives}{True \,Positives\,+\,False \,Negatives}\) |
False positive rate (specificity) | Proportion of negative samples mistakenly predicted as positive, in relation to all negative samples | \(\frac{False\, Positives}{False\, Positives\,+\,True \,Negatives}\) |
False negative rate | Proportion of positive samples mistakenly predicted as negative, in relation to all relevant samples, i.e., all positive samples | \(\frac{False \,Negative}{False \,Negative\,+\,True \,Positive}\) |