Weka Tutorial 33: Random Undersampling (Class Imbalance Problem) Video Lecture | Weka Tutorial - Data & Analytics

hi this is where tutorial 33 and in this
tutorial I'm going to demonstrate how
you can do random undersampling on your
data set so I'm going to use a filter
and that filter will will make your
imbalance data set balanced by randomly
under sampling the majority class
remember that I had a tutorial where I
showed you how you can synthetically
produce instances for the minority or
the majority class but this tutorial is
a bit different than that tutorial you
can find the link here with the
difference is very simple that in smooth
you are using an algorithm to to
generate instances for the minority
class or the majority class mostly you
generate instances for minority class
using small so that the class becomes
almost balanced but in this tutorial I'm
going to randomly under sample the
majority class so that the number of
instances for one class is it becomes
equal to the number of instances in
other class so in this tutorial I'm just
assuming that my dataset contains only
two classes so to do with to demonstrate
with this I'm going to explore and I'm
going to open a file this is my
customized file spy from my research
project but you can use any other files
so here if I if you if you carefully
take a look at the class it's very
imbalanced because we have two classes
yes and no nes county is just 372 and
frequency of know is about 6,500 so it's
a big difference between the bedrug and
minority class here so what I'm going to
do is I'm going to under simple the
majority class here which is class no
and I'm going to take that six thousand
four hundred and forty nine instances
under sample so that they be can the
number of negative instances here
becomes exactly the same as the
instances in
the positive class or ES class which is
372 so in that case I'm going to apply a
filter that will reduce the oversample
class here the major sorry not over
sample the major ad class here so that
the number of instances of the blow both
of the classes they become equal so in
order to do that I'm going to the filter
and in I'm going to supervise filter and
attributes sorry I'm going to deal with
instances you see this mode is there so
we have dealt with its mode previously
to generate synthetic synthetic
instances for different classes but here
I'm going to use the split sub simple
filter and if you click on the options
then you can see that you have you have
an option called distribution spread and
to make them equal you have to make you
have to put one there in the
distribution spread I click OK and apply
a filter and then you can see that the
number of class instances are never
exactly same exactly equal to each other
so number of years is 372 I did not do
anything on the years class it remained
it remains 372 but is from 6400 49
instances of no classes I just randomly
under sample that class so that it is in
the number of instances for these
classes again it is also 372 so now
you're having a balanced class problem
rather than having imbalanced class
problem so hope that helps thank you
very much
keep watching my tutorials I'm coming
with more tutorials and also don't
forget to subscribe and put a click on
like button if you if you're if you
really got help from this tutorial
thanks a lot