Interactive ROC Curves

Kyle Yang

The Receiver Operating Characteristic (ROC) curve is a way to assess the accuracy for the outcomes of a binary test.
ROC curve is very useful to evaluate the model when the observations are balanced between each class. I often review the definition of ROC curves because it is easy for me to forget what x, y axes stand for and why the plot is like this.
So here, with the help of the open source project (credit to Gagarine Yaikhom), I built this interactive visualization to help people like myself understand and remember the details of ROC curves.

Assume there are 1000 actual positives and 1000 actual negatives. The red curve shows distribution of negatives and the green curve shows distribution of positives. This distribution is obtained from the result of a classifier which estimates the probability of a test point being positive.
The curve (AUC) is generated because the rate of change for true positives and false positives are different.
Hover over the threshold line in the distribution groups plot and move it, and then ROC curve and the table will be adjusted accordingly.