Weka Tutorial 06 Discretization (Data Preprocessing) Video Lecture - Weka

hello in this tutorial I'm going to show
you how you can use the discretized
option in the wake of machine learning
workbench so what's the discretization
when you have selected features from our
previous tutorial we have seen that we
have some feature values and they are
scattered really scattered from they
range from minus 2.5 4 for example to
infinity or sometimes there are no
values for example na n but for some
classifying algorithms this has been
proved that they work well when you
discretize your values feature values so
what what is a discretization it means
that you are going to use a predefined
set of intervals and you are going to
group your feature values according to
those the interval values so that's what
discretization means so in this tutorial
I'm going to show you how you can use
the discretize a filter in Vega machine
learning workbench so in this tutorial
I'm going to use the diabetes dot K rff
file and you can see that we have 99
features in this particular ARF file and
we have some features called Prius and
so on I don't know the meaning of the of
those feature names but we have to just
stick with this example and our ninth
feature is our classification feature
either date it classifies those
instances either as positive instances
or negative instances so you can see
that if you select any other attributes
here for example number 2 class you can
see that we have the selected attribute
pane provides you the information about
this particular attribute and we have
minimum values maximum values mean of
those values and standard deviation and
so on so what what I'm going to do here
is I'm going to classify that with
classifying algorithm called naive Bayes
I'm using 10-fold cross-validation and I
click start so we have 76 76 % if
measured as a weighted average and this
has been proved that naive Bayes works
well when you discretize your values
your feature values that means you have
scattered feature values and now you're
going to group them according some
according to some predefined
set of interval values so you know to do
that you'll have to select the filter we
can filter and supervised filter
attribute filter and then you select
discretize and you have plenty of
options like this attributing in this s
inverse selections and so on but I'm
going to use the default one and I'm
leaving you guys with these other
options to play with so for this
tutorial six sick I'm going to stick
with only the default setup and I
applied discretize and you can see that
things have changed a lot now if I
select the number two attribute called
class you can see that instead of mean
and standard deviation minimum and
maximum values for your particular data
set you have some ranges for example
minus infinity to 99.5 you in this range
you have 197 instances and 150 4.5 to
infinity positive infinity
you have 122 instances in your data set
for this particular attribute called
plus and you have this for other
instances or for other features as well
so now I'm going to do is I'm going to
use the same classifier again standard
10-fold cross-validation with naive
Bayes and you can see that the F measure
has jumped to seventy seven point nine
from 76 this was the previous result 76
percent F measure and this is the latest
result seven seven one nine percent so
that means that nine base particularly
works well for this particular dataset
if you use discrete discretization
rather than using scattered values of
features