Weka Tutorial 29: Precision-Recall Curve (Model Evaluation) Video Lecture | Weka Tutorial - Data & Analytics

Oh welcome to another wicked tutorial in
this tutorial I'm going to show you how
you can produce a precision recall car
that's a really really wonderful and
useful visualization tool to see the
classifier performance so in the last
tutorial I've shown you how you can use
the ROC curve to see the performance of
your classifier and most of the times in
the literature the PR carves or the
precision recall car and the ROC curves
are really really useful tool so right
after the ROC curve I'm just going to
show you how you can use the PR curve as
well so I'm using the same or file as
before as in the last tutorial and you
see so diabetes.org file can be found in
the wake up data repositories and this
our file is having tested negative and
tested positive two classes there so to
produce the ROC curve sorry to produce
the PR curve I'm going to use the nine
based classifier to classify and on the
results list again sorry it on the
classifier output you can see that
costing sensitive analyzes the precision
recall net measure so precision recall
curve is sitting in the middle of cost
sensitive and costing census insensitive
a lysis of your classifier performance
but it's still more to the cost
sensitive part so in order to produce PR
curve we just right click on this result
in the result list you can go to the
visualize special curve just like the
ROC curve tutorial and you can see that
we're having two options there because
our diabetes.org file had two classes
this is negative and tested positive so
we're just interested in tested negative
if you click that you can see that we're
having a ROC curve here but we're not
interested in ROC curve in this tutorial
we were interested in ROC curve in the
last tutorial so in order to produce a
precision recall curve your x-axis is
going to
to be recall our true positive rate okay
and your y axis is going to be precision
so that's your PR car for this
classifier you can see well you will see
that this this PR car is that is not a
bad one it's not a good one but to me
it's more to the good part because in PR
car so this is the position axis and
this is the recoil axis so one one is a
design is the desired graph to be
produced from any classifier any good
classifier so if your classifier is more
to the right upper right corner so that
classifier is the best classifier you
are having that's the perfect classifier
and this is not this cannot be Oh
this can be seen in many cases but
usually this doesn't happen in most of
the empirical analysis you are not
having a perfect PR car for your
classifier so you can see that in this
in this PR this PR curve is more to the
upper right corner and this is not a
perfect one but this is not a bad one at
all
so this PR curve is a really good one so
more to the upper right corner your PR
curve is the more the better the
classifier is so that's the conclusion
so I hope that helps if you really like
my tutorials don't hesitate to put click
on the like button and also try to
subscribe on my channel because I'm
trying to be very frequent in the last
few in the next few months because I'm
getting some relief from my PhD work so
I'll be trying my best to produce more
away car tutorials on different
different topics so don't forget to
subscribe don't forget to like put likes
laughs for my tutorials and thank you
very much next tutorial I'm going to
cover you cover how you can compare
multiple numbers of rock cars produced
from the
different types of classifiers thank you
very much