Custom Ensemble Machine Learning Algorithm for Classification Problem

[Omprakash_Jena]

Hello Everyone!

Trust you are doing good.

I want to create a custom ensemble ML algorithm for a classification problem. Say for example Random Forest is an ensemble algorithm and decision trees are used to create the same.

I want to create a custom ensemble model with combination of say decision tree, random forest, gradient boosted trees, naive bias. Finally use voting for prediction.

Is this possible in KNIME? Any suggestions would be really helpful.

Many Thanks!
Omprakash Jena

[ScottF]

Hi there -

Take a look at the top branch of this example workflow, which gives you an idea of how you can combine PMML models of various types into an ensemble:

Spark Compiled Model Predictor – knime

You may also need to occasionally make use of the Tree Ensemble Model Extract for certain nodes like the Random Forest Learner, as shown here:

Training a Decision Tree – rs

Does that help?

[Omprakash_Jena]

Thanks for the quick response, @ScottF ! This would be really helpful if could explain the workflow for me. I am getting confused with this. Not sure how to use the voting method while prediction.

Your support is really helpful.

Many Thanks!
Omprakash Jena

[ScottF]

In the first workflow, maybe just focus on the initial nodes:

The two learner nodes are creating models based on the training data, and the PMML to Cell nodes convert those into KNIME tabular format. Those two tables are then stacked together with the Concatenate so that you have both models in the same table. Then the Table to PMML Ensemble converts them to a single model.

You could use a PMML Predictor downstream, along with the test data, to apply your new ensemble model. Does that make sense?

[Omprakash_Jena]

Thanks a lot, @ScottF ! Make sense.

Regards,
Omprakash Jena