Prevalence
| {p | watch tp fn} / ({p | watch tp fn} + {n | watch fp tn}) = |
{prevalence | watch tp tn fp fn | percent} |
Balance of classes |
Sensitivity
| {tp} / {p | watch tp fn} = |
{sensitivity | watch tp fn | percent} |
True positive rate, recall, hit rate, statistical power |
Specificity
| {tn} / {n | watch tn fp} = |
{specificity | watch tn fp | percent} |
True negative rate, selectivity, confidence level |
Informedness
|
{sensitivity | watch tp fn | percent} +
{specificity | watch tn fp | percent} -
100% =
|
{informedness | watch tn fp tp fn | percent} |
Youden's J, bookmaker informedness, Peirce’s I |
Positive predictive value
| {tp} / {predp | watch tp fp} = |
{positive_predictive_value | watch tp fp | percent} |
Precision |
Negative predictive value
| {tn} / {predn | watch tn fn} = |
{negative_predictive_value | watch tn fn | percent} |
|
Markedness
|
{positive_predictive_value | watch tp fp | percent} +
{negative_predictive_value | watch tn fn | percent} -
100% =
|
{markedness | watch tp fp tn fn | percent} |
Δp ('delta-p') |
Matthews correl. coef.
|
√({informedness | watch tn fp tp fn | percent} × {markedness | watch tp fp tn fn | percent})
=
|
{matthews_cc | watch tp fp tn fn | percent} |
φ ('phi') |
False positive rate
|
{fp} / {n | watch tn fp} =
|
{false_positive_rate | watch tn fp | percent} |
Fall-out, 100% - {specificity | watch tn fp | percent}, significance level |
False negative rate
|
{fn} / {p | watch tp fn} =
|
{false_negative_rate | watch tp fn | percent} |
Miss rate, 100% - {sensitivity | watch tp fn | percent} |
False discovery rate
| {fp} / {predp | watch tp fp} = |
{false_discovery_rate | watch tp fp | percent} |
100% - {positive_predictive_value | watch tp fp | percent} |
False omission rate
| {fn} / {predn | watch fn tn} = |
{false_omission_rate | watch fn tn | percent} |
100% - {negative_predictive_value | watch tn fn | percent} |
'Accuracy'
| ({tp} + {tn}) / {total | watch tp tn fp fn} = |
{accuracy | watch tn fp tp fn | percent} |
Fraction correct |
F1 score
|
( 2 × {positive_predictive_value | watch tp fp | percent_no_small_digits} × {sensitivity | watch tp fn | percent_no_small_digits} ) / ( {positive_predictive_value | watch tp fp | percent_no_small_digits} + {sensitivity | watch tp fn | percent_no_small_digits} )
=
|
{f1_score | watch tn fp tp fn | percent} |
Sørensen–Dice coefficient |