best way to reduce features in knime


i have a dataset containing 200000 records with 100 attributes.......

what is the best way to reduc feature in this dataset?

i would appreciate your help

any one?

Hello m.a.najimi:

Depends on what you need to do. If your goal is a classification task, you can use one of the nodes weka coming annexed to KNIME (meta). To do the above you should download and use Weka AttributeSelected Classifier. To reduce dimensions You could also use Principal Component Analysis (PCA) of KNIME. That will serve to both classification and regression tasks and for clustering analysis.

I hope you serve


Hi m.a.najimi,

A part of reducing dimensions by projecting all your features into its principal components (PCs) as suggested by Gabriel (this is maybe the best option), you can also select the best features by using the Feature Elimination metanode.

The Feature Elimination metanode implements a feature elimination method using a backward strategy. You need only to connect the metanode, substitute the learning and predicting nodes by the ones of your choice, set the Backward Feature Elimination Filter criteria, and that's it!

I hope this help.



We just published a white paper about feature reduction:

thanks alot man...

is it possible to download the workflows in the paper?

Thank you Iris,

very useful indeed.