kindly Confirm me the following steps of feature selection using weka.
1: open file and choose dataset.
2: click on select attribute
3: choose attribute evaluate and search method
4: selection of attribute method. if i used only full training set then is it ok or not.
5: start
thanks
Hi,
If you are woking with train data, what about testing data.
Is testing data is % of split of train data or performing cross-fold validation?
If yes once you select the attributes as per requirement, no need to specify in next step (i.e, in classify tab).
else need to give same attribute selection for test data if mentioning separately the test data.
Bests
Shareef
first of all thanks Mr. Shareef,
i have just used training option during feature selection. after choosing features then i have apply classifier using 10 fold cross validation. Is it correct or should i used cross fold during feature selection also. plz guide.
The above steps are correct.
About your secind question, first select the suitable features alone then using k fold cross validation to train your training data.
Hi,
To correct and conclude:
Frankly, if your aim is performing attribute selection and classifying the selected subset after that, then the "Select attribute" panel is not the right place to do that.
On the other hand, cross-validation in "Select attribute" panel does not prevent overfitting. All it does is giving you an indication on how each attribute is chosen when the attribute selection method is running on several features subsets (the training sets are used in cross-validation, while the test sets in this cross-validation are not used at all).
HTH.
Samer
This video promotes a wrong implimentation of feature selection using WEKA. This is because feature selection and classification are not evaluated properly in one process. In other words, the selected features may not be used to build the final classification model. Hence, being a member at WEKA environment, I can say that such video is not recommended.
HTH.
Samer
Hi Samer,
In the above post you have mentioned that as select attribute panel is not correct to do so. Can you please guide me the right way to perform attribute selection followed by classification of subsets in weka.
Thanks in advance
Priya
Hi Priya,
The most reliable way is using the meta-learner "AttributSelectedClassifer".
Cheers,
Samer, PhD.
For selection of the feature i have done following methods. Please correct if i performed wrong
1. i took all features and got some result say 85-90% using KNN.
2. i observed the feature there it contain 0 in last three features so i removed those three features and got result same as in step 1. i.e., 85-90% using KNN
3. Then i have used weka attribute selector it selected some features and i have applied KNN classifier and got 83%.
Now i am in dilemma that which one i should consider for declaring the result.
Pls help me in this regard.
Satish,
I'm not sure how did you perform feature selection and classification, but both procedures have to be accomplished under the evaluation of cross-validation to assess that the best features are used by the KNN algorithm to build the final predictive model.
Cheers,
Dr. Samer Sarsam
Thanks Dr. Samer Sarsam ,
i have set 10 fold cross-validation for my dataset to divide the train and test.
With Regards,
SATISH
You're welcome, Satish.
If your aim is building a classification model based on a selected feature subset, you need to select and then classify your data *together* from the "Classify" tab only. The 2 procedures have therefore to be done under the evaluation of cross-validation.
Regards,
Dr. Samer Sarsam
Thank you Sir, now i looking forward to your valuable suggestion.
Thanks once again
With regards,
SATISH
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining