Weka Tutorial 13: Stacking Multiple Classifiers (Classification) Video Lecture | Weka Tutorial - Data & Analytics

hello and welcome to the way Commission
learning tutorial in this tutorial I'm
going to show you the way how you can
use wake up in order to combine multiple
classifiers to classify training or test
instances in machine learning problem
nowadays this is very popular to use
multiple classifiers instead of one
single classifier
there are however advantages and
disadvantages of using or combining
multiple classifiers as an advantage we
can say that instead of one classifier
algorithms power we are currently using
two classification algorithms power so
the model we will induce by combining
those two those two classifiers or
multiple transceivers
will be more reliable and will be more
sophisticated to identify or classify
instances from the training set of cross
validation set or the test set however
there are some disadvantages one
overfitting problem many research papers
have stated that if you combine multiple
classifiers in order to classify
instances from your data set that time
you are actually prone to get
overfitting problem to the
classification algorithms have their own
drawbacks so when you are combining more
than one classification algorithms in
your test at that time you are at the
risk to have the multiple classification
algorithms problems in one go but I'm
just leaving you guys to get some
research in depth reading on using or
combining multiple classifiers in order
to get or in order to induce a
classification model in this tutorial
I'm just simply showing that Rekha is
powerful in
have to provide you the facility for
combining more than one classification
algorithms for inducing a single model
now I will tell you one thing the
combination of multiple multiple
classifiers for instance classification
this task is actually following three
types the first type is called in
symbols so what is instantly and simply
means simply if you are using two
algorithms for example begging and
boosting begging and boosting are
stand-alone classification algorithms
but they use the way they work of their
methodologies they use or combine
multiple classification algorithms tasks
into one so whenever you are using n
symbols that means you are actually
using either bagging or boosting the
second one is called voting and the
third one is called stacking so I'm just
leaving you guys to know exactly what
voting and stacking perform when you are
using on a particular data set for
combining multiple classifiers so now
I'm going to show you how you can use
multiple classifiers how can you use an
assembly approach voting approach and
stacking approach in order to do that we
will use the wiki Explorer we're going
to open a file and we are browsing to
the wake up folder
on data folder and we're choosing
silence data set so this is our iron
status F so straight away I'm jumping to
classify time here the first one first n
symbols so in classifier option you can
choose and then you can go to meta and
here you have two options
the first one is edibles m1 this is a
boosting algorithm and the second one is
bagging
this is bagging outlet of the average or
normal or trivial bagging algorithm so
if you use either of adaboost m1 or
begging at that time you actually enjoy
following the approach called in sibling
you are actually combining multiple
classifiers because inside edibles and
bagging we have the methods of more than
one classification algorithms so in in
order to do that you can just click on
bagging and you start classifications
okay again you can choose adaboost and
use the classification now we are
interested on in voting and stacking so
because voting and stacking in Weka the
choose the selecting of voting or
stacking is almost the same the
parameters are also the same I'm just
showing you one in this case which is
voting so we go down to meta and we find
nested dichotomies and here at the end
we have vote and also you can see that
we have the option stacking as well so
we click on vote so now we go up and we
click on this n you can see in
classifier step we have one way cut
classifiers dot classifier
this means that currently we are just
having one default classifier
but we are actually trying to combine
multiple classifiers so we click on here
you can see that we have an option
banner
the default is set to zero are for this
particular tutorial we are not
interested in zero are so we are going
to delete that from the option again we
go to choose and for this task I'm going
to select naive Bayes so this is my
first classifier to add to the pool the
second classifier I am going to combine
with naive Bayes is Bayesian eight so
I'm selecting that and we run clicking
add so now at this point if you want to
insert more classifiers into this pool
you just click on choose button and
choose the desired classification
algorithms and click on add so if I
close this window you can see that now
we have two waked classifiers dot
classifier the combination rule I'm just
leaving
you guys with this option to play with
because there are multiple combination
rules several research papers talked
about this and you should find the
suitable combination rule for your task
but we have some options like average of
probabilities product out probabilities
median etc so we are just keeping for
the sake of this tutorial
default one average of probabilities
okay if you want to know more click on
more button and if you want to know the
capabilities then you can click on
capabilities these options are always
open for you to dig into so I'm clicking
ok and if I click on start this means
that I am using the voting approach by
using by combining two classification
algorithms one is nine base the other
one is a snake
the on the basis of the combination rule
average of probabilities so I click on
start and the classification task is
done