average_rocdata: average_rocdata

In AndrewLawrence/dCVnet: doubly cross-validated elastic-net regularised generalised linear models

average_rocdata

Description

Function calculates the average of a set of logistic regression ROC curves. For a range of thresholds between 0 and 1 the sensitivity and specificity are extracted and mean averaged over the set of curves. This is intended to be used with k-fold cross-validation data.

Usage

average_rocdata(rocdata, n = 1000)

Arguments

rocdata	a extract_rocdata object.
n	the number of thresholds in \[0,1\] to evaluate.

Details

Warning: The averaging of roc-curves is difficult
    (e.g. \href{https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4112395/}{
    Chen & Samuelson Br J. Radiol Aug 2014})
    , particularly while preserving the AUROC such that the AUROC
    for the average ROC-curve is equal to the average AUROC of the
    component ROC-curves.

    This function is for a rough display of average performance
    over bootstraps or repeats of k-fold cross-validation.

    Minimal testing suggests that there should be agreement in the
    AUROCs to the third decimal place.

Value

a data.frame object of class "rocdata" which can be plotted. Contents:

• Sens : Sensitivity

• InvSpec : 1 - Specificity

• alpha : threshold

• run : a label indicating this is averaged

See Also

plot.rocdata, extract_rocdata

AndrewLawrence/dCVnet documentation built on Sept. 24, 2024, 5:24 a.m.