A decision boundary separates regions assigned to different predicted classes. In binary classification, logistic regression produces probabilities from 0 to 1, then applies a threshold to turn those estimates into class predictions. section 5.5
Use the boundary as an operational choice rather than treating it as fixed:
- Compare possible thresholds with precision-recall and receiver operating characteristic curves. section 5.5
- Check whether predicted probabilities are calibrated with reliability diagrams, root mean squared calibration error, or Brier score. section 5.5
- Adjust the operating point with Platt scaling, isotonic regression, or Youden’s index when appropriate. section 5.5
Cytochrome P450 inhibition
For predicting whether a compound inhibits a metabolic enzyme, logistic regression can estimate a probability and a chosen decision boundary can convert it into an inhibition or non-inhibition prediction. The boundary affects how predictions are evaluated and used in drug–drug interaction work. section 5.5






