Classification with imbalanced datasets question

Follow the full discussion on Reddit.
I've been working on a medical classification project with an imbalanced tabular dataset. I have 3 classes, and each class has 44, 16, and 14 rows of data respectively. When I train a random forest classifier, I see that my model is only predicting the dominant class for all test instances most of the time. How can I get around to this? Also, are there any recommendations you can give me for dealing with imbalanced datasets? Thank you