I have built a random forest model for classification on a certain data set where the final classes are ordinal in nature, i.e. bad, average and good. From the theoretical knowledge, few of the variables are known to have fixed slopes in the final prediction, i.e. if their values increase the predicted class or at least the probabilities should only increase in a specific direction. But due to the non-linear nature of random forest, these are not showing up properly in the final model. I was wondering, is there any specific variation of random forest which I should try or if there is some other model where I can fix the direction like this for the model?
Sumit Das thanks for your answer, but I have gone through the link earlier. And the link contains a very generic improvement approach for random forest. My question is more specific and therefore the link does not help much. What I wanted to know was in a random forest, is there any way to fix the directional effect of certain variables.
Mohammed Deeb thanks for your answer. The first paper seems to directly address my question.
