Exam Professional Machine Learning Engineer topic 1 question 159 discussion - ExamTopics

Problem:

An online message board uses a machine learning model to detect toxic comments. The model shows bias, incorrectly flagging non-toxic comments from underrepresented religious groups. This leads to false positives and user complaints.

Constraints:

The team has limited budget and is already overextended.

Possible Solutions:

• A. Add synthetic data: Generate training data showing the targeted phrases used in non-toxic contexts to retrain the model.
• B. Replace with human moderation: Remove the automated system and revert to manual review of all comments.
• C. Replace with different classifier: Switch to an alternative text classification model.
• D. Raise the threshold: Adjust the model's sensitivity, requiring a higher toxicity score before flagging a comment.

Suggested Solution:

The suggested solution is to add synthetic training data (A), which addresses the bias without requiring a complete model overhaul or incurring significant additional costs.