# Negative Precision and Recall Curve

I am evaluating a classification model, Support Vector Machine, and I am having difficulty interpreting the Precision and Recall Curve 1 graph.

For example, this graph plots a straight negative line. I wasn't expecting this, especially when the accuracy of this model is 0.99, and it has a false negative rate of almost 0, and I also don't have false positives. That is, true positives are at 100% and true negatives are at 99.99%. You can check this info in the confusion matrix 2. How do I interpret a graph like this? Is my model completely wrong?

## 1 Answer

A generally negative slope is to be expected in precision-recall space. When you set the decision threshold below zero, you classify everything as positive, so your recall is 1.0 and your precision is equal to the prevalence of the positive class*; when you set the decision threshold close to one, you classify nearly nothing as positive, so your recall approaches 0.0 and your precision is hopefully close to 1.0.

*That suggests a mismatch between the plot and your expectations: the plot appears to be treating "Working" as the positive class (note that $$3485/(3485+13)\approx0.9963$$ matches the lower-right point of the plot), while you seem to be treating "Malfunctioning" as the positive class ("I don't have false positives").

Lastly, it is odd that the plot should be a straight line. Perhaps your SVM has lots of ties (or you're not passing in the confidence scores, just the hard predictions?)?