how to compar AUC when it give negative value, I use trapz(x,y) to calculate it

Q: when I calculate the area under curve using trapz function some of them was negative value ( so I thought parts below the axis are negative and those above are positive), I know because it is area we remove the sign. however in case I want to compare this auc with another auc, does I keep the sign or just take the absolute value ? thanks Amirah

[Matt J]

How did you end up with an ROC curve with negative values?

[the cyclist]

By construction, the AUC cannot be negative. For example, in the attached image, the AUC is the area between the blue line and the x-axis. Even if the blue line is below the "random model" curve (the diagonal), it will have positive AUC.

Please supply some more detail about your results.

[Eldhose Poulose]

Can I say AUC of a coin flip be Linear and always Positive?

[the cyclist]

Do you mean if you flip a coin a number of times, and try to predict the outcome?

AUC is a number. It is the area under the curve -- not the curve itself. That number will always be positive.

Do you mean will the ROC curve be linear? On average, the curve will be linear from the bottom left to the upper right. But for any given experiment, it will deviate from that line, due to sampling error.

Answer 1

[Harshan Ravi]

Its the trapz function expects the input. The function requires vector in x axis in an increasing order and their values in corresponding y-axis. Eventhough, the order of the x and y doesnot matter for plotting, it is important for trapz.