Train a classification model using the Decision Tree algorithm. Evaluate the accuracy of the class prediction by scoring metrics, ROC Curve, and Lift Chart.
This is a companion discussion topic for the original entry at https://kni.me/w/wWrebA_HNv4hHDDG
Great job @Maarit !
Finally we have a simple brand new workflow showing how to score classification models!
Thanks @paolotamag! Yes - those interactive views are nice on their own like this, but even more powerful when combined with other views in components. I am currently working on more of these examples, and they will all be shared here via the Workflow Hub!
Related to this topic of evaluating model performance.
- testing classification models (different types or just different features etc) to see which perform the best on the data you have
- would like to be able to connect multiple Predictors into the Scorer… and the Scorer retains the input table name or Node name… that way the Lift and ROC curve can be plotted on one graph so you can see performance across models
You can currently do this with the new Binary Classification Inspector node - check out some of the example workflows featuring it on the Hub!
I have tried using the feature scorer but it does not allow to connect to more than one predictor and so am unable to compare. please guide. Thanks.
I think there might be some language confusion going on here, as I’m not really sure what you mean by the question.
Is your question about comparing the results of different models, each with their own predictor? In that case you can get the scoring metrics for each model, and combine them into a single table for comparison.
But neither case requires a scorer node with multiple inputs.
Thanks very much.
I am referring to the model comparison report that has been generated above.
That report is from a software package other than KNIME, I’m not sure what exactly. I believe @DemandEngineer was using it as an example of the kind of thing he’d like to produce.
I presumed thay the report was generated using Knime and hence the query.
Please ignore my question and Thanks again for the very prompt responses.