Data & Analytics Exam > Data & Analytics Videos > Weka Tutorial > Weka Tutorial 36: Learning Curve 1 (Model Evaluation)
Weka Tutorial 36 Learning Curve 1 (Model Evaluation) Video Lecture - Weka
Text Transcript from Video
by in this tutorial we're going to learn
how to produce learning curves using
weapon so learning curves are really
important aspects in in machine learning
because using learning curves you can
easily identify whether the learner you
have generated it is overfitting or not
so there are two different ways to
produce learning curves so we are going
to cover one in this tutorial and the
next will next tutorial will show you
how to do the second option to produce
learning curves so the first option to
produce learning curve is to very data
for example if you have 1,000 instances
in your data set then you are going to
produce a learner from 10% of this data
and it produces an error rate so you
have the error rate for 10 percent of
the data in the next run you can produce
you can just give the classifier twenty
percent of your data and you have the
error rate for that twenty percent of
the data and these in this way you can
it can go up to ninety percent of the
data you're fitting to your classifier
and you record the error rate for the 90
percent of the data and then on the
x-axis you are having the percentage of
the data you fit to your classifier ten
percent to ninety percent and on the
y-axis you have the error rates for the
for the respective amount of data and
you just plot the error rate and that's
your learning curve so in this tutorial
I'm going to show you how you can do
that how you can produce learning curves
just by varying the amount of data
instances in your data set so this is
kind of tricky but bear with me so we
are going to choose experimental we
cannot do it on Explorer so we're going
to choose experimenter and this is
interesting to know that period
previously we were sticking with the
simple configuration mode in
experimenter but this time we have to go
to advanced so we select advanced and we
have to click new because we're trying
to create a new experiment
so now a bunch of options have popped up
now from these options you can see that
first option is destination this is this
is a minor thing this is not significant
because in destination you're just
putting the file where you're going to
save all of your results produced from
this experiment so you can have you can
save them in CSV you can save them in
your database that I've never done and
you can also save them in your AR file
so I can select instances results
listener and I just click to edit the
properties of this option and again I'm
selecting the output file by left
clicking it and you can see that I can
go to desktop and save it as 1 dot R so
this is the route not really significant
for this tutorial we just click on OK
now the important thing here is the
result generator so we have an option
here we click Choose you can see lots of
option here but you if you are following
my tutorials then you know that I'm a
cross-validation guy so that's my
favorite
I'm just saved I'm just selecting
cross-validation result producer and as
this option is selected now I'm going to
left click on it to edit the properties
we left click on it and this editor
opens we have numbers of information and
options there number of foals output
files raw output we are not interested
in them we are interested in split
evaluator option now in split evaluator
we can choose several options if you are
if you're trying to create learning
curve for a regression algorithm then
you can select regression split
evaluator but in this experiment we're
just trying to produce learning curves
for an eye based classifier which is a
classification algorithm so we are drawn
we are selecting classifier split
evaluator but you can see that you have
other two options here as well if you
want to know more about them then you
can select the more button here now as
you have selected the classifier split
evaluator we left click on this to edit
the properties again a property editor
pops up
and you have several options there
attribute ID class for ir statistics and
print argot column and you are not
interested in them so we're just trying
to select the classifier here and as I
told you that we're trying to produce
learning curves for 90 based classifier
we can select nine based classifier
directly from here but I'm not going to
do that because what we are trying to do
here is we are trying to systematically
increase the amount of data to be fed to
now based classifier so in that case we
have to choose a meta classifier and
from the meta classifier we are
selecting filteredclassifier and now if
we left click on this filteredclassifier
option then we can select our classifier
called ninety base and as a filter it's
not going to be discretized though it's
it's the default there but if you click
on choose then several filter options
popped up and from these options you can
select the wakeup unsupervised instance
filter called remove percentage because
we are going to remove 10 percent 20
percent up to 90 percent of the data and
we're going to feed them to 9 based
classifier to see how much error rate it
is producing for each of the individual
percentage of data so now you can click
OK click OK again and again so now it's
done so runs are from 1 to 10 that's
fine
in generator properties it's disabled by
default so you have to enable that and
then you have an option popping up just
like this and in this option expand
split evaluator then you have another
option called classifier expanded to and
you have an option called filter expand
that and you can see that you have
percentage under filter you select that
and click on the select button now you
can see something came up on this window
50 so this is the percentage of the data
you are going to feed to your night
based classifier
for example in our learning curve we are
going to put nine error rates one for
10% the second for 20% and ninth for 90%
of the data okay so we are just going to
delete this one and add 10 20 and this
goes up to 90 so I have given 9
intervals here ranging from 10% to 90%
and this amount of data will be provided
to our night based classifier it will
have error rates for 10% 20% up to 90%
and then we're going to plot them this
is the idea behind this and as data set
we are going to click on add new and
from Wake our data repository we are
going to select the iris our file we are
clicking on re start off our file and
click on open now we're going to go to
run tab and we just click on start so
you can see that our process has started
and finished we didn't we don't have any
errors we go to analyze tab that's ok
and we click on experiment now from this
comparison field here you select the
percentage incorrect which which is the
error rate synonymous so we click on
perform test and you can see that we
have nine error rates produced from the
iris dataset and the first one here 4.4
you can see that in this case our minus
P is 10 point zero this means that it's
error rate is 4.4 when we have excluded
10% of data from our dataset so the
error rate of 90 base is 4.4 when it was
working with 90 percent of the data we
just took out 10 percent of it now the
second column here is four point four
seven this means that the 90 based
classifiers error rate is four
4:7 when we took out twenty percent of
the data randomly from our dataset okay
so these are the error rates for the
Kenai Bayes classifier for ten percent
to 90 percent interval of the data that
are taken out from the original data set
now if we plot them in an excel file so
I have the excel file here so we have so
these are these another data these are
the data taken out okay so these are the
data taken out from the original data
set ten percent taken out nine base
already is 4.4 twenty percent taken out
my base error rate is four point four
seven and we only have ten percent of
the data that is 90 percent taken out
their rate is 15 point six seven so if
you if you select them go to insert and
include a scatterplot here you can see
this is the learning curve for our night
based classifier as we have varied the
in amount of instances in our dataset
from 10 percent to 90 percent as we have
removed this amount of data from our
data set so from this learning curve
with naked eye you can see that the
night based classifier that we have
produced on iris dataset it performs the
best with 90 percent of the data in the
data set and it performs the worst when
we have taken most of the data we are
just keeping ten percent of the data for
its training and test
I mean cross-validation so you can see
you can say that the nine based
classifier you have produced performs
the based when the when you have when
you are keeping most of the data in your
data set as we have varied instances
number of instances in our data set so I
think and hope and believe that this was
helpful for you to produce learning
curves using wake up in the next
tutorial it's going to be a follow up of
this tutorial where we are going to see
how you can produce learning curves by
not varying the data but varying
different parameters of your algorithm
thank you very much
how to produce learning curves using
weapon so learning curves are really
important aspects in in machine learning
because using learning curves you can
easily identify whether the learner you
have generated it is overfitting or not
so there are two different ways to
produce learning curves so we are going
to cover one in this tutorial and the
next will next tutorial will show you
how to do the second option to produce
learning curves so the first option to
produce learning curve is to very data
for example if you have 1,000 instances
in your data set then you are going to
produce a learner from 10% of this data
and it produces an error rate so you
have the error rate for 10 percent of
the data in the next run you can produce
you can just give the classifier twenty
percent of your data and you have the
error rate for that twenty percent of
the data and these in this way you can
it can go up to ninety percent of the
data you're fitting to your classifier
and you record the error rate for the 90
percent of the data and then on the
x-axis you are having the percentage of
the data you fit to your classifier ten
percent to ninety percent and on the
y-axis you have the error rates for the
for the respective amount of data and
you just plot the error rate and that's
your learning curve so in this tutorial
I'm going to show you how you can do
that how you can produce learning curves
just by varying the amount of data
instances in your data set so this is
kind of tricky but bear with me so we
are going to choose experimental we
cannot do it on Explorer so we're going
to choose experimenter and this is
interesting to know that period
previously we were sticking with the
simple configuration mode in
experimenter but this time we have to go
to advanced so we select advanced and we
have to click new because we're trying
to create a new experiment
so now a bunch of options have popped up
now from these options you can see that
first option is destination this is this
is a minor thing this is not significant
because in destination you're just
putting the file where you're going to
save all of your results produced from
this experiment so you can have you can
save them in CSV you can save them in
your database that I've never done and
you can also save them in your AR file
so I can select instances results
listener and I just click to edit the
properties of this option and again I'm
selecting the output file by left
clicking it and you can see that I can
go to desktop and save it as 1 dot R so
this is the route not really significant
for this tutorial we just click on OK
now the important thing here is the
result generator so we have an option
here we click Choose you can see lots of
option here but you if you are following
my tutorials then you know that I'm a
cross-validation guy so that's my
favorite
I'm just saved I'm just selecting
cross-validation result producer and as
this option is selected now I'm going to
left click on it to edit the properties
we left click on it and this editor
opens we have numbers of information and
options there number of foals output
files raw output we are not interested
in them we are interested in split
evaluator option now in split evaluator
we can choose several options if you are
if you're trying to create learning
curve for a regression algorithm then
you can select regression split
evaluator but in this experiment we're
just trying to produce learning curves
for an eye based classifier which is a
classification algorithm so we are drawn
we are selecting classifier split
evaluator but you can see that you have
other two options here as well if you
want to know more about them then you
can select the more button here now as
you have selected the classifier split
evaluator we left click on this to edit
the properties again a property editor
pops up
and you have several options there
attribute ID class for ir statistics and
print argot column and you are not
interested in them so we're just trying
to select the classifier here and as I
told you that we're trying to produce
learning curves for 90 based classifier
we can select nine based classifier
directly from here but I'm not going to
do that because what we are trying to do
here is we are trying to systematically
increase the amount of data to be fed to
now based classifier so in that case we
have to choose a meta classifier and
from the meta classifier we are
selecting filteredclassifier and now if
we left click on this filteredclassifier
option then we can select our classifier
called ninety base and as a filter it's
not going to be discretized though it's
it's the default there but if you click
on choose then several filter options
popped up and from these options you can
select the wakeup unsupervised instance
filter called remove percentage because
we are going to remove 10 percent 20
percent up to 90 percent of the data and
we're going to feed them to 9 based
classifier to see how much error rate it
is producing for each of the individual
percentage of data so now you can click
OK click OK again and again so now it's
done so runs are from 1 to 10 that's
fine
in generator properties it's disabled by
default so you have to enable that and
then you have an option popping up just
like this and in this option expand
split evaluator then you have another
option called classifier expanded to and
you have an option called filter expand
that and you can see that you have
percentage under filter you select that
and click on the select button now you
can see something came up on this window
50 so this is the percentage of the data
you are going to feed to your night
based classifier
for example in our learning curve we are
going to put nine error rates one for
10% the second for 20% and ninth for 90%
of the data okay so we are just going to
delete this one and add 10 20 and this
goes up to 90 so I have given 9
intervals here ranging from 10% to 90%
and this amount of data will be provided
to our night based classifier it will
have error rates for 10% 20% up to 90%
and then we're going to plot them this
is the idea behind this and as data set
we are going to click on add new and
from Wake our data repository we are
going to select the iris our file we are
clicking on re start off our file and
click on open now we're going to go to
run tab and we just click on start so
you can see that our process has started
and finished we didn't we don't have any
errors we go to analyze tab that's ok
and we click on experiment now from this
comparison field here you select the
percentage incorrect which which is the
error rate synonymous so we click on
perform test and you can see that we
have nine error rates produced from the
iris dataset and the first one here 4.4
you can see that in this case our minus
P is 10 point zero this means that it's
error rate is 4.4 when we have excluded
10% of data from our dataset so the
error rate of 90 base is 4.4 when it was
working with 90 percent of the data we
just took out 10 percent of it now the
second column here is four point four
seven this means that the 90 based
classifiers error rate is four
4:7 when we took out twenty percent of
the data randomly from our dataset okay
so these are the error rates for the
Kenai Bayes classifier for ten percent
to 90 percent interval of the data that
are taken out from the original data set
now if we plot them in an excel file so
I have the excel file here so we have so
these are these another data these are
the data taken out okay so these are the
data taken out from the original data
set ten percent taken out nine base
already is 4.4 twenty percent taken out
my base error rate is four point four
seven and we only have ten percent of
the data that is 90 percent taken out
their rate is 15 point six seven so if
you if you select them go to insert and
include a scatterplot here you can see
this is the learning curve for our night
based classifier as we have varied the
in amount of instances in our dataset
from 10 percent to 90 percent as we have
removed this amount of data from our
data set so from this learning curve
with naked eye you can see that the
night based classifier that we have
produced on iris dataset it performs the
best with 90 percent of the data in the
data set and it performs the worst when
we have taken most of the data we are
just keeping ten percent of the data for
its training and test
I mean cross-validation so you can see
you can say that the nine based
classifier you have produced performs
the based when the when you have when
you are keeping most of the data in your
data set as we have varied instances
number of instances in our data set so I
think and hope and believe that this was
helpful for you to produce learning
curves using wake up in the next
tutorial it's going to be a follow up of
this tutorial where we are going to see
how you can produce learning curves by
not varying the data but varying
different parameters of your algorithm
thank you very much
About this Video
4.69/5 Rating
Apr 29, 2026 Last updated
Related Exams
Video Description: Weka Tutorial 36: Learning Curve 1 (Model Evaluation) for Data & Analytics 2026 is part of Weka Tutorial preparation. The notes and questions for Weka Tutorial 36: Learning Curve 1 (Model Evaluation) have been prepared according to the Data & Analytics exam syllabus. Information about Weka Tutorial 36: Learning Curve 1 (Model Evaluation) covers all important topics for Data & Analytics 2026 Exam. Find important definitions, questions, notes, meanings, examples, exercises and tests below for Weka Tutorial 36: Learning Curve 1 (Model Evaluation).
Introduction of Weka Tutorial 36: Learning Curve 1 (Model Evaluation) in English is available as part of our Weka Tutorial for Data & Analytics & Weka Tutorial 36: Learning Curve 1 (Model Evaluation) in Hindi for Weka Tutorial course. Download more important topics related with notes, lectures and mock test series for Data & Analytics Exam by signing up for free.
Description
Video: Weka Tutorial 36: Learning Curve 1 (Model Evaluation) of Weka Tutorial have been curated by the Data & Analytics experts, helping you revise the topic quickly for exam preparation. Watch on EduRev.
Information about Weka Tutorial 36: Learning Curve 1 (Model Evaluation)
Here you can find the meaning of Weka Tutorial 36: Learning Curve 1 (Model Evaluation) defined & explained in the simplest way possible. Besides explaining types of Weka Tutorial 36: Learning Curve 1 (Model Evaluation) theory, EduRev gives you an ample number of questions to practice Weka Tutorial 36: Learning Curve 1 (Model Evaluation) tests, examples and also practice Data & Analytics tests.
Related Searches
practice quizzes, pdf , Viva Questions, study material, Objective type Questions, Important questions, Summary, shortcuts and tricks, Sample Paper, Semester Notes, Weka Tutorial 36: Learning Curve 1 (Model Evaluation), past year papers, Weka Tutorial 36: Learning Curve 1 (Model Evaluation), video lectures, Previous Year Questions with Solutions, Exam, Weka Tutorial 36: Learning Curve 1 (Model Evaluation), ppt, Extra Questions, Free, MCQs, mock tests for examination;