I am trying to predict the suffix a certain word will take based on properties of the word. There three outcomes but they are not absolute, that is, some words can take two or three different suffixes (but importantly, some can only take one):

A w1

A w2

B w3

C w3

C w4


I am trying to predict this, including the probability at which a given word with multiple suffixes will take each of those suffixes. Which algorithm can handle these type of task the best?

So far I have tried random forest (with cforest), ndl (with the ndl package) and a simple neural network with (with multinom from the nnet package), but they all seem to perform quite poorly when predicting probabilities directly (brier scores and mse scores are really high).

I've gotten the best performance if I set some sort of threshold for the predicted probabilities (at about 0.3), mark those as 0, and for anything above that as 1. This leaves me with a result of "can take the suffix or not", which is not optimal but sort of works...

Is there a better approach?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.