ROC curve under diagonal?

Question

I trained an SVM to classify images based on some extracted features (using the ISIC dataset). The resulting ROC curve produced by sklearn looks like this:

I have don't quite understand the line for class 0 and 1 (orange and green). I have never seen a ROC curve under the diagonal. What does this tell me now and is this even a valid curve? The other computed evaluation metrices are the following if that helps my question.

Answer 1

The curve is valid.

• The most commonly seen ROC curves have a curve above the diagonal, which indicates that large predicted scores are associated with the label, and low predicted scores are not associated with the label.
• The diagonal, i.e. the line $\text{TPR}=\text{FPR}$ of a ROC curve corresponds to the model that has a completely random relationship predicted scores and labels.
• A curve below the diagonal is a model that as predictions which are opposite the labels. Low predicted scores are associated with the label, and high predicted scores are not. This is the opposite of the first case.

For more information about how ROC curves are constructed, see

• How to calculate Area Under the Curve (AUC), or the c-statistic, by hand
• Understanding ROC curve
• ROC curve crossing the diagonal