How can I combine the result of SVN with a different classification method?

MorSha

I need to create a prediciton model.

The file contains 5 columns:

- choice (label - polynominal)

- column_a (polynominal)

- column_b (real values)

- column_c (real values)

- column_d (real values)

I tested several combination of MODELS and I got the best results using BAGGING of SVN LibSVN (without using the ploynominl feature because SVN cannot use it)

Even if the accuracy seems good, I believe adding the missing feature should improve the results.

I tired to use VOTE and STACKING to combine results from SVN with another classifier model, but the problem is I cannot do it because the SVN does not include all the features.

How can I combine the result of SVN with a different classification method?

jacobcybulski

All models in a Stacking ensemble must use the same label and the same attributes, so the best way to use a mix of models is to ensure they have the same pre-processing before hand, e.g. convert nominal attributes to numerical to please SVM or Logistic Regression, while Decision Tree will still perform well on transformed attributes.

sara20

@MorSha,

Hello

There are some great links for your question you can take a look on them

https://academy.rapidminer.com/learn/article/ensemble-methods

https://docs.rapidminer.com/latest/studio/operators/modeling/predictive/ensembles/bagging.html

https://docs.rapidminer.com/latest/studio/operators/modeling/predictive/ensembles/stacking.html

Best
Sara

Telcontar120

Why not simply convert your polynominal attribute to numerical using dummy coding (Nominal to Numerical operator).  This should enable it to be used by your SVM model and any other model types that require numerical attributes.