MLA-C01 Question 150
Select 2You are training a machine learning model using Amazon SageMaker to predict customer churn. After initial training, the model's performance is lower than expected, as indicated by a high validation loss. The dataset is unbalanced, with only 10% of the samples representing churn cases. Which of the following actions should you take to refine the model and improve its performance?
- A
Use a weighted loss function to handle class imbalance.
- B
Perform feature scaling on the input data to normalize its distribution.
- C
Increase the batch size during training to process more data at once.
- D
Generate synthetic samples for the minority class using techniques like SMOTE.
- E
Switch to a more complex model architecture without changing the dataset.
- F
Use cross-validation to assess the model's performance more effectively.
Show answer and explanation
Correct answers: A, D
Explanation
The high validation loss is likely caused by the class imbalance in the dataset. To refine the model, you need to address the imbalance by using techniques such as a weighted loss function or generating synthetic samples for the minority class. These approaches improve the model's ability to correctly predict churn cases. Other options, like feature scaling or increasing batch size, do not directly target the issue and are less likely to improve performance.
- A. Correct.
Using a weighted loss function is a valid technique to address class imbalance by giving more importance to the minority class during model training.
- B. Incorrect.
Feature scaling is a general preprocessing step, but it does not directly address the issue of class imbalance, which is the main problem in this scenario.
- C. Incorrect.
Increasing the batch size might improve training efficiency, but it does not address the class imbalance leading to poor model performance.
- D. Correct.
Generating synthetic samples for the minority class using techniques like SMOTE can help balance the dataset and improve the model's performance on the minority class.
- E. Incorrect.
Switching to a more complex model architecture without addressing the underlying data imbalance may lead to overfitting or failing to improve performance on the minority class.
- F. Incorrect.
Cross-validation is useful for evaluating model performance but does not directly address the problem of class imbalance or improve the model.