Random Tree implementation always predicts one label

asked 2017-03-29 06:59:09 -0500


I use RTrees in order to achieve multi-label classification in RF signal IQ samples. The issue that I'm facing is that my predictor always returns a specific label, without taking in mind the other available classes. If I open the trained .xml model with a text editor I can see different nodes to be labeled with a variaty of classes.

In addition, the trained model's test error is reported to be around 20% and the train error sometimes is reported as zero.

Does this behavior rings any bell?

Thanks in advance, Kostis

edit retag flag offensive close merge delete