You are troubleshooting a custom text classification project in Language Studio where your model's macro F1-score is low. Which THREE remedial actions can help improve the macro F1-score across all categories? (Choose two.)
Trap 1: Manually hardcode all possible classification outputs in client…
Hardcoding defeats the purpose of machine learning classification.
Trap 2: Delete all training data across every category.
Deleting all data destroys the model.
- A
Manually hardcode all possible classification outputs in client application code.
Why wrong: Hardcoding defeats the purpose of machine learning classification.
- B
Review and correct inconsistent or ambiguous labels in the training dataset.
Correcting label noise improves model discrimination.
- C
Delete all training data across every category.
Why wrong: Deleting all data destroys the model.
- D
Ensure a balanced distribution of training sample counts across categories where possible.
Balancing class distributions prevents bias toward majority classes.
- E
Add more representative training documents for underperforming categories.
Adding data for weak classes improves recall and F1-score.