How to Work Around Multinomial Regression's Reference Level Issue Without a Natural Baseline.

Using Predict() with Multinomial Distribution Models: A Solution for Class Probabilities in GBM