Can we use SMOTE in H2O models?

[somnathdas76]

I am new to KNIME. Is there any way by which we can use SMOTE in H2O models?

[mlauber71]

@somnathdas76 you can use the smote node to balance your training data. You might take a look at the threads linked in this workflow under external resources.

Use SMOTE and ROSE algorithms to balance data – mlauber71

You can use the balanced data for your model training. You might want to keep the balancing in mind when you interpret your results with your test data.

Some H2O nodes (in KNIME) offer a generic balancing option. You might have to see if this can help you.

https://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/algo-params/balance_classes.html

Note that the resulting model will also correct the final probabilities (“undo the sampling”) using a monotonic transform, so the predicted probabilities of the first model will differ from a second model.

You might also tell us more about your data and case. Maybe other options like XGBoost might also help (with an evaluation statistic like AUCPR).