Feature Selection using Maxent classifier

Question

I am using maxent classifier on a dataset with 138,314 rows and 43 attributes. The form of training data are like below:

     Feature             Class
     Cloudy Sad... Humid Outdoor

If I test the model on held-out data, the accuracy is just above 70%, I want to improve accuracy by feature selection. My question is, how to perform feature selection using Maxent classifier? Is there other ways to improve accuracy in Maxent model?

Or like Ratnaparkhi A paper said:

We use a very simple feature selection strategy: assume that any feature that occurs less than 5 times is noisy and discard it.(Ratnaparkhi, A. (1998))

Any help will be appreciated.