Skip to content

Identification of relevant thresholds for binarisation #4391

Open
@csudre

Description

@csudre

Is your feature request related to a problem? Please describe.
Currently all the confusion matrix based metrics ask for an already binarized input of prediction with a threshold predecided. In order to calculate metrics when the chosen binarisation threshold provides a given (other) confusion metric value, there should be methods linked between transform:post and confusion matrix so as to define that threshold and then calculate the relevant confusion matrix metrics

Describe the solution you'd like
A function to find the relevant threshold at which to binarise to reach a certain value of confusion matrix and then calculate the associated other confusion metrics values at that threshold. Can be derived from the calculation of ROC, Precision/Recall curve or FPROC. This is necessary for the following:

  • PPV@Sensitivity
  • Sensitivity@PPV
  • Specificity@Sensitivity
  • Sensitivity@Specificity
  • Sensitivity@FPPI

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions