calcAUC {ebdbNet} | R Documentation |
Returns the approximate Area Under the Curve (AUC) of a Receiver Operating Characteristic (ROC) curve.
calcAUC(sens, cspec)
sens |
Vector of sensitivity values, calculated for varying thresholds |
cspec |
Vector of complementary specificity values, calculated for the same varying thresholds as |
Let TP, FP, TN, and FN represent the number of true positives, false positives, true negatives and false negatives of inferred network edges, respectively. Sensitivity is defined as
\frac{TP}{TP + FN}
and complementary specificity is defined as
\frac{TN}{TN + FP}
Note that sens
and cspc
should be in the same order with respect to the threshold value
so that their elements correspond. That is, if the first element of sens
was calculated at a
threshold value of 0.01 and the second at a threshold value of 0.02, then the first element of cpsec
should be also be calculated at a threshold value of 0.01 and the second at a threshold value of 0.02, and
so on. The AUC is approximated using the trapezoid method, and can take real values between 0 and 1. An
AUC of 0.5 indicates a classifier with random performance, and an AUC of 1 indicates a classifer with
perfect performance.
AUC of the ROC curve
Andrea Rau
library(ebdbNet)
tmp <- runif(1) ## Initialize random number generator
## Generate artificial values for sensitivity and complementary specificity.
fn <- function(x) {return(-1/(x^7)+1)}
set.seed(1459)
sens <- c(fn(seq(1, 2.7, length = 100)),1) ## Sensitivity
cspec <- seq(0, 1, by = 0.01) ## Complementary specificity
## Calculate the AUC of the ROC curve
AUC <- calcAUC(sens, cspec) ## AUC of this ROC curve is 0.9030868