MLS-C01 Question 222
Select 3You are building a machine learning model on AWS SageMaker to predict customer churn. During model evaluation, you notice that the training accuracy is very high, but the validation accuracy is significantly lower. Which strategies can you use to address this issue?
- A
Increase the size of the training dataset by incorporating more labeled examples.
- B
Add regularization techniques such as L2 regularization to the model.
- C
Increase the number of epochs during training to allow the model to learn better.
- D
Use cross-validation to tune hyperparameters and improve generalization.
- E
Reduce the feature set by removing features with low importance.
Show answer and explanation
Correct answers: A, B, D
Explanation
The scenario describes overfitting, where the model performs well on the training data but poorly on validation data. To address this, strategies such as increasing the size of the training dataset, adding regularization techniques, and using cross-validation for hyperparameter tuning are effective. These approaches improve the model's ability to generalize to unseen data. Increasing epochs or arbitrarily reducing the feature set without analysis may not address the root cause of overfitting and could even worsen the issue.
- A. Correct.
Increasing the size of the training dataset can help reduce overfitting by providing the model with more diverse examples, making it generalize better to unseen data.
- B. Correct.
Adding regularization, such as L2 regularization, penalizes overly complex models, preventing them from overfitting to the training data.
- C. Incorrect.
Increasing the number of epochs can lead to overfitting, as the model may memorize the training data rather than learning generalizable patterns.
- D. Correct.
Using cross-validation helps tune hyperparameters, ensuring the model generalizes well to unseen data and avoids overfitting or underfitting.
- E. Incorrect.
Reducing the feature set is not guaranteed to prevent overfitting unless the removed features are explicitly causing noise or redundancy in the model.