ensemble learning based on learners created using different data


I am currently wondering if it is possible in Knime to create multiple models using differnt sample data having the same class, then use enseble type method to increase the predictive performance. For example

- Model/Learner 1 is trained using Attribute 1, 2 and 3 with class X

- Model/Learner 2 is trained using Attributes 4, 5 and 6 with the same class X

- The outcome of model 1 and model 2 are weighted and compared to give a better prediction for X.

Hi Paulio,


yes this is possible. first you therefore need the Ensemble Plugin. (http://www.knime.org/ensemble-learning)

With it you can for example transform the pmml into a cell. Combine those cell with a Table to pmml ensemble node and use the pmml ensemble predictor for the final prediction. The workflow is attached.

Here is a paper about this: http://www.inf.uni-konstanz.de/gk/pubsys/publishedFiles/FiAdGa13.pdf

and a blog post here https://www.knime.org/blog/pmml-ensembles-and-knime

Best, Iris

Hi Thanks Iris,

This is great.