This workflow demonstrates the usage of the Feature Elimination Meta Node. The first input is labeled training data, the second test data which does not need to be labeled. Inside the meta node at least a learner and a predictor node need to be inserted between the loop start and the loop end node. Alternatively a partitioning node or even a cross validation meta node can be used. The loop iterates over all columns and iteratively removes the attribute that has the lowest influence on classification accuracy. After the loop has been finished the filter node can be used to filter out all attributes that do not affect classification accuracy much.
This is a companion discussion topic for the original entry at https://kni.me/w/zAVBDgFn4GZfiv2K