MLS-C01 Question 88
Select 2An e-commerce company uses AWS SageMaker to train a machine learning model for predicting customer purchase behavior. However, during testing, the model demonstrates significant bias against a particular customer demographic. Which of the following mitigation strategies could help address the bias issue?
- A
Collect and include more diverse training data that represents all customer demographics equally.
- B
Use SageMaker's built-in feature importance functionality to remove sensitive features like age and gender from the dataset.
- C
Implement SageMaker Clarify to analyze and detect bias in the dataset and model predictions.
- D
Increase the size of the training dataset without analyzing its distribution to improve model performance.
- E
Fine-tune the model hyperparameters to reduce training loss and improve generalization.
Show answer and explanation
Correct answers: A, C
Explanation
Bias in machine learning models often originates from imbalanced or non-representative training data. Including diverse training data and using tools like SageMaker Clarify to detect and mitigate bias are effective strategies to address this issue. Simply increasing the dataset size or adjusting hyperparameters does not resolve the underlying problem of bias, and removing sensitive features may not eliminate bias due to feature correlations.
- A. Correct.
Including diverse training data ensures that the model is exposed to a balanced representation of all customer demographics, which reduces bias.
- B. Incorrect.
Removing sensitive features like age or gender might not be the best strategy because bias can still emerge from correlated features. Instead, bias detection and mitigation tools like SageMaker Clarify should be used.
- C. Correct.
SageMaker Clarify is specifically designed to detect and measure bias in datasets and models, helping to identify and mitigate potential fairness issues.
- D. Incorrect.
Increasing the size of the training dataset without analyzing its distribution might improve model performance slightly, but it does not guarantee that the bias will be reduced if the dataset remains imbalanced.
- E. Incorrect.
Fine-tuning hyperparameters can improve the model's accuracy but does not directly address the issue of bias in the dataset or model predictions.