The caret package for R provides a variety of error metrics for regression models and 2-class classification models, but only calculates Accuracy and Kappa for multi-class models. Therefore, I wrote the following function to allow
caret:::train to calculate a wide variety of error metrics for multi-class problems: