One big problem with the PMML format however are the random forest models. Indeed, due to the complexity of the random forest models - including multiple decision trees, each with slightly different input features and training – there is currently no universal consensus on how the PMML format should represent random forest. The result is that every company delivers its own PMML format or just waits till times are ripe.
The KNIME Analytics Platform (www.knime.org) is no exception. The Tree Ensemble Learner node that implements the learner node for the random forest algorithm (http://www.dataminingreporting.com/blog/decision-tree-ensemble-decision-tree-forest) does not produce a PMML-formatted model, but just a KNIME-formatted model.
All hopes are not lost though!
You can use the Tree Ensemble Model Extract node to extract the model parameters into an XML-formatted cell. At this point the XML-formatted cell containing the model can be read into a PMML model using the Table to PMML Ensemble node.
There! Now you have your random forest model in the shape of a PMML model!