Hi Mate, By using AUC ROC curve tells about only TPR rate and AUC of the model it also tells about the threshold value of the model as well right… As per my understanding in the plot whenever the curve goes decreasing from point 2 to point 3 then at that particular point at Y axis is the best threshold value of the model is this right… Thanks,
@Rajesh B Yes, the point when the slope starts decreasing, it can work as a good threshold. But in real-world, you won't find smooth curves and the slope won't always decrease. So, it is better to choose the threshold by looking at the (TPR, FPR) pairs. If you find a suitable (TPR, FPR) according to your problem statement then choose the corresponding threshold. You can watch this video to know how to find a threshold of a point and area under curve: th-cam.com/video/TEkvKx2tQHU/w-d-xo.html
@@NormalizedNerd Thanks for the reply... But how to choose the best threshold for a model between 0 to 1 we will get many confusion matrix's there will be code for choosing best threshold.
The line below the yellow line is not worse than random prediction because you can flip its prediction 😉 although surely it's a bug! Always random prediction is the worst case scenario
Good One. Short and sweet. You made the topic soo easy with the example.
excellent , you cleared all my doubt
Superb clarification
Hi Mate,
By using AUC ROC curve tells about only TPR rate and AUC of the model it also tells about the threshold value of the model as well right…
As per my understanding in the plot whenever the curve goes decreasing from point 2 to point 3 then at that particular point at Y axis is the best threshold value of the model is this right…
Thanks,
@Rajesh B Yes, the point when the slope starts decreasing, it can work as a good threshold. But in real-world, you won't find smooth curves and the slope won't always decrease. So, it is better to choose the threshold by looking at the (TPR, FPR) pairs. If you find a suitable (TPR, FPR) according to your problem statement then choose the corresponding threshold.
You can watch this video to know how to find a threshold of a point and area under curve:
th-cam.com/video/TEkvKx2tQHU/w-d-xo.html
@@NormalizedNerd Thanks for the reply...
But how to choose the best threshold for a model between 0 to 1 we will get many confusion matrix's there will be code for choosing best threshold.
Sir very good explanation, keep making videos like this god bless you.
Thank you, I will!
Thank you very much!!
Hi, video was helpful but you can use a larger cursor.
Thank you...suggestion noted :)
Excellent thanks
Glad you liked it
Literally bingewatching your vvideos right now and I am actually not interested in Machine Learning at all :D
The line below the yellow line is not worse than random prediction because you can flip its prediction 😉 although surely it's a bug! Always random prediction is the worst case scenario