using c45 and naive bayes for feature selection

-3

I want to apply the following algorithm for my feature selection:

Shuffle the training data and take a 10% sample.
Run C4.5 on data from step 1.
Select a set of attributes that appear only in the first 3 levels of the simplified decision tree as relevant features.
Repeat 5 times (step 1-3)
Form a union of all the attributes from the 5 rounds.
Run Naïve Bayesian classifier on the training and test data using only the final features selected in step 5

I am pretty new from weka.. so I would like to know how can can select a set of attributes that appear in the first 3 level. What classes should I use?

feature-selection weka

asked Mar 09 '11 at 13:22

Alex Hernandez
40●8●10●15

3 Answers:

oldest newest most voted

I've tried similar things with Weka's J48 classifier (it's implementation of C4.5). Unfortunately, this quite a pain. The only way I found was via the command line using the -g option, which outputs a graph representation of the classifier. e.g.

java -cp /usr/share/java/weka.jar weka.classifiers.trees.J48 -g -t training.arff -o -d model.dat

digraph J48Tree {
N0 [label="married" ]
N0->N1 [label="= YES"]
N1 [label="YES (6.0/2.0)" shape=box style=filled ]
N0->N2 [label="= NO"]
N2 [label="NO (12.0/4.0)" shape=box style=filled ]
}

You could capture this text and parse it to get the node names.

answered Mar 09 '11 at 21:17

Cerin
402●25●37●44

edited Mar 09 '11 at 21:17

3	I would suggest you post such Weka-specific questions to the Weka mailing list (preferably after looking briefly at the documentation, which you can find at the Weka home page). answered Mar 09 '11 at 15:04 Oscar Täckström 1459●10●27●43

this algorithm works pretty nice for the feature selection. but unfortunately there is no direct way of performing the same in weka. if you want to run the same indirectly then you can perform following steps: 1. load your dataset. 2. go to classifier selection and select C45 means J48 from the tree based classifier list. 3. perform the classification. 4. tree model will be generated and will be shown in the right side output panel. 5. from the generated tree find out the first three level of attributes. 6. now load the same dataset again. 7. use the filter remove from the attribute.unsupervised list and remove the attributes which are not in you list which you had generated in the step 5. 8. now classify the data with any of the other classifier that you want to use.

i hope this will be helpful to you...

regards, Ankit.

answered Jul 28 '11 at 01:10

ankit
1●1●1●1

Your answer

toggle preview

community wiki

Subscription:

Once you sign in you will be able to subscribe for any updates here

Tags:

feature-selection ×27
weka ×10

Asked: Mar 09 '11 at 13:22

Seen: 1,016 times

Last updated: Jul 28 '11 at 01:10

using c45 and naive bayes for feature selection

Subscription:

Related questions