Home / Questions / 349954
Accepted
Juan Luis
Asked: 2020-04-28 05:43:47 +0800 CST 2020-04-28 05:43:47 +0800 CST

How to improve ROC curve plot in R?

  • 772

I have a binomial logit model and I am evaluating the goodness of fit of my model using the ROC curve :

library(pROC)
curva_ROC <- roc(modelo_logit_viv$y,modelo_logit_viv$fitted.values)
plot(curva_ROC,col = "blue",xlim = c(1,0),ylim = c(0,1),
     xlab = "Especifidad", ylab ="Sensibilidad", main = "Curva de ROC")
auc(curva_ROC) # 0.8445

This is the graph that returns me:

ROC curve

I want to modify it as follows:

  • Delimit the x-axis (specificity) in such a way that the beginning is almost at 1 and ends at 0, that is, that the limits of the horizontal axis coincide with the diagonal (I thought that by indicating I xlab = c(0,1)would obtain that result, but the graph does not vary when include this argument in the plot function).

  • That the area between the curve and the diagonal is colored in a lighter blue than the curve.

  • That the value of the area appears auc(curva_ROC) right in the center of said area.

r

1 Answers

  1. Best Answer

    1. Adjust axes

    The way to do it in the base plots is through the parameters xlimand ylimI couldn't verify it with your data, but it works properly with the example I put together for this answer. On the other hand xlab, they ylabonly establish the "labels" or labels of each axis

    2. Color area

    The behavior of plot.ROC()supports coloring the entire area of ​​the curve, using these parameters (example):

    auc.polygon = TRUE 
auc.polygon.col=rgb(.35,0.31,0.61, alpha = 0.4)

    What you are looking for, however, is to color the curve all the way to the diagonal. One way could be to generate the polygon below the curve and up to the diagonal, taking advantage of the data of theplot.ROC()

    polygon(with(p, cbind(specificities, sensitivities)), 
        col = rgb(.25,0.31,0.61, alpha = 0.4), 
        border = "blue",
        lwd = 2)

    3. Text with AUC value

    You can configure and adjust it with the following parameters:

    print.auc = TRUE,
print.auc.x = 0.73,
print.auc.y = 0.6

    The values ​​of xe yshould be adjusted so that the text fits inside the area.

    Example

    library(pROC)

curva_ROC = roc(aSAH$outcome, aSAH$s100b, levels=c("Good", "Poor"))
p <- plot(curva_ROC,
          col = "blue",
          print.auc = TRUE,
          print.auc.x = 0.73,
          print.auc.y = 0.6, 
          xlim = c(1,0),
          ylim = c(0,1),
          xlab = "Especifidad", 
          ylab ="Sensibilidad", 
          main = "Curva de ROC")

polygon(with(p, cbind(specificities, sensitivities)), 
        col = rgb(.25,0.31,0.61, alpha = 0.4), 
        border = "blue",
        lwd = 2)

    Result:

    enter image description here

    Sources:

    • 1