SOCR logoBPAD

Classification Metrics Explorer

Move prevalence, sensitivity and specificity independently and watch predictive values, accuracy and F1 respond.

  • sensitivity
  • specificity
  • predictive value
  • prevalence

Ranking metrics (threshold-free)

AUC
0.695
Brier score
0.107
Log loss
0.357
Prevalence
0.127

Distribution of predicted probabilities

threshold00.10.20.30.40.50.60.70.80.910510152025303540Predicted probability of malignancyCount
BenignMalignant

Confusion matrix at current threshold

Pred +Pred −
True +019
True −0131

Threshold-dependent metrics

AUC, Brier score and log loss never move as you drag the slider — they summarize the whole ranking, not one operating point.
MetricValueType
Accuracy0.873Threshold-dependent
Sensitivity0.000Threshold-dependent
Specificity1.000Threshold-dependent
PPV0.000Threshold-dependent
NPV0.873Threshold-dependent
Youden's J0.000Threshold-dependent
AUC0.695Ranking / probability
Brier score0.107Ranking / probability
Log loss0.357Ranking / probability
Why 0.5 is rarely the right threshold
Drag the threshold toward 0.2 and watch sensitivity climb while PPV falls — moving the line only trades one error type for the other along a fixed probability ranking. With prevalence near 0.13, accuracy can look good even from a model that is not clinically useful, which is exactly why predictive values must always be read together with the underlying prevalence.