"Multi-class Classification using SVM"

D_M

Hi,

I want to classify data into positive, negative & neutral.

For this I want to use 2 SVM one trained on positive & not positive data & the other on negative & not negative data. Now I want to take the data predicted as positive by the 1st classifier & not negative by the second classifier as positive & data predicted as negative by the 2nd classifier & not positive by the 1st classifier as negative. In rest of the cases I want to tag the data as neutral.

Is it possible to do this in Rapid-Miner in GUI Mode?

fischer

Hi D.M.,

please look at the AttributeConstruction operator. You can use the if(condition,thenValue,elseValue) function to construct your target attribute based on the two other predictions.

Best,
Simon

D_M

Thanks for replying.

I am very new to rapidminer. Can u plez explain a bit more? I am not able to put the 2 classiifers under the same root process.  Can you please tell how can I put the 2 classifers under the same root process?

fischer

Dear D.M.,

that depends very much on how your data looks like and what you want to achieve. A general approach would be the following:

- Use an AttributeConstruction to generate "isPositiv" and "isNegative" binary attribute
- Declare the first as a label (ChangeAttributeRole)
- Train a model
- Set the second as a label
- Train another model
- Then, apply both models subsequently to the test set, and perform the inverse attribute construction on the two prediction attributes you get.

I would recommend to have a look at the RapidMiner training courses. E.g., an introductory course is this one: http://rapid-i.com/content/view/110/143/lang,en/

Best,
Simon