MLA-C01 Question 220
Select 3You are training a machine learning model using Amazon SageMaker to predict customer churn. During evaluation, you notice that the model performs very well on the training dataset but poorly on the validation dataset. Which of the following methods can help identify whether the model is overfitting or underfitting?
- A
Analyze the training and validation loss curves during training
- B
Increase the size of the training dataset and retrain the model
- C
Check the model's performance metrics on a test dataset
- D
Regularize the model by adding dropout layers and re-evaluate
- E
Visualize the model's predictions on the training and validation datasets
Show answer and explanation
Correct answers: A, C, E
Explanation
To identify model overfitting or underfitting, you should analyze the training and validation loss curves to observe discrepancies, check performance metrics on a test dataset to evaluate generalization, and visualize predictions to assess whether the model is capturing meaningful patterns. These methods provide insights into the model's behavior and help pinpoint overfitting or underfitting issues.
- A. Correct.
Analyzing the training and validation loss curves during training can help identify overfitting or underfitting. For example, if the training loss is much lower than the validation loss, it is an indicator of overfitting.
- B. Incorrect.
While increasing the size of the training dataset is a technique to mitigate overfitting, it does not directly help in identifying whether the model is overfitting or underfitting.
- C. Correct.
Checking the model's performance metrics on a test dataset can validate whether the model generalizes well and help identify potential overfitting or underfitting.
- D. Incorrect.
Adding dropout layers and re-evaluating the model is a method to reduce overfitting, but it does not help in identifying overfitting or underfitting directly.
- E. Correct.
Visualizing the model's predictions on the training and validation datasets can help identify patterns that indicate overfitting or underfitting, such as whether the model is failing to generalize to unseen data.