I'm quite new with KNIME. I recently took Data Mining course and my faculty gave me project on "GERMAN DATA SET"
I ran SVM(Support Vector Machine) without any feature subset selection on my dataset. I've built a project with the classical nodes (file reader, partitioning , SVM learner,SVM predictor, scorer) and executed. My accuracy is coming around 75%.
I tried Low variance filter and PCA for feature subset selection but my accuracy is decreasing rather than increasing.
Can you suggest any Feature Subset Selection mechanism by which i can increase my accuracy?
Other methods for increasing accuracy will be highly appreciated?
Did you have a look at the metanode? It has a short description in it. The input is the dataset the output depends on which features you have selected in the feature elimination filter node.
The second output is the second input but only the relevant columns (which you have selected inside the metanode). You can use the second output to train any model (ideally the same as you used in the feature elimination loop) and use KNIME's standard learner-predictor-scorer schema. The metanode itself does not do the final prediction or scoring.