MLS-C01 Question 221
Select 3You are building a deep learning model on Amazon SageMaker to predict customer churn. After training the model, you notice that the validation loss is significantly higher than the training loss. Which of the following approaches can help address this issue?
- A
Add regularization techniques such as L1 or L2 regularization to the model.
- B
Increase the number of training epochs to ensure the model learns better.
- C
Reduce the number of features by performing feature selection or dimensionality reduction.
- D
Increase the size of the training dataset by using data augmentation or collecting more data.
- E
Enable dropout layers during training to prevent the model from relying too heavily on specific neurons.
Show answer and explanation
Correct answers: A, D, E
Explanation
The validation loss being significantly higher than the training loss is a classic symptom of overfitting. To address this, you can use regularization techniques such as L1/L2 regularization, increase the training data size to improve generalization, or use dropout layers to prevent reliance on specific neurons. Simply increasing training epochs or reducing features without considering the underlying issue may not solve the problem and can sometimes worsen it.
- A. Correct.
Adding regularization techniques like L1 or L2 helps to penalize overly complex models, which can reduce overfitting and improve generalization.
- B. Incorrect.
Increasing the number of training epochs without addressing the root cause of overfitting may worsen the problem by allowing the model to memorize the training data.
- C. Incorrect.
Reducing the number of features may help with overfitting in some cases, but it is not guaranteed to address the issue unless the features are irrelevant or redundant.
- D. Correct.
Increasing the size of the training dataset helps the model generalize better by exposing it to more diverse examples, which reduces overfitting.
- E. Correct.
Dropout layers randomly disable neurons during training, which helps prevent the model from being overly reliant on specific patterns or neurons, reducing overfitting.