I am working on intent classification task (chatbot engine), 2k sentences, 24 classes.
Major class is composed of about 150 sentences, minor class of about 35 sentences, the others are more or less balanced (70 sentences each one).
I used Fast-text pre trained embedding and then feed into CNN using Keras. I've done a Grid search cross validation to choose best parameter for the model, then i trained model with these parameters on train (SMOTE applied to balance classes) and then evaluate on test (hold-out).
I got an accuracy and micro avg F-score of 0.91
What are other good metrics to evaluate performance of model?
Confusion metrics? ROC curve? classification report?
I tried also Precision-recall obtained the results below in the image.
How should be the values for each class to be considered good?