WELCOME Abdennour : Software engineer

Apr 18, 2012

Lab1: Classification (Supervised): (With Weka)

In this lab you will learn about using weka using the database iris.arff you find it  in the folder: ~ / weka-3-6/data/Iris.arff.
This is a basic donnnées, very famous, with 150 examples of flowers described by four attributes of continuing value and belonging to three classes.

1.DATA :
1>Open firstly file Iris.arff with a text editor to discover the format ARFF (Attribute Relation File Format) :

2>Click on the button "O". Choose the data file. / Data / Iris.arff:

Some information will appear in the window

In the pane "SelectedAttribute", you can get basic statistics for the selected attribute: Name, Type, Missing, Unique, distinct, Vlaur Min / Max.

Select SepalLength Attribut:

Select SepalWidth ATTribut  : 

Select PetalLength :

Select PetalWidth :

Select Class:

So we have : 
     >>>Blue Color => Iris-Setosa
     >>>Red Color => Iris-versicolor
     >>>Last Color => Iris-virginica

-Click on Visualize All : 

==>We Note that the overlap(chevauchement(fr))  between classes is minimal when we have classification based on attribute petal (Either  petalLength or petalWidth).

-Weka offers the possibility to operate by pretreatment by applying a filter on attributes:
- Click on Choose Button >supervised>attribute>Discretize :
==>This filter allows to discretize continuous values

2.Data visualization :

For a first approach to classification, go in the "Visualize" Tab. You will see a set of 25 graphics:

1. Change the axes to achieve a classification that gives off the lower of decision rules

-When X-> PetalWidth && Y -> SepalLength , We give off a minimal number of  Decisions Rules : 

Decisions Rules (Règles de décision(fr)) :
 -if(X>=0.1) AND(X<=0.6) Then Class<-- Iris-Setosa.
-if(X>=1) AND(X<=1.7)  AND(Y<=5.6) Then Class<-- Iris-Versicolor.
-if(X>=1.8) AND(X<=2.5)  Then Class<-- Iris-virginica.
-if(Y>=7.1)  Then Class<-- Iris-virginica.

=>This is the best classification in comparison with other classifications. In effect, the overlap between classes is minimal

1 comment: