MLA-C01 Question 119
Select 4You are building a machine learning model using Amazon SageMaker to predict customer churn for a subscription service. The dataset has significant class imbalance, where only 5% of the customers in the dataset have churned. Which of the following steps should you take during model development to address this imbalance and improve the performance of your model?
- A
Use oversampling techniques to balance the dataset by increasing the number of churned samples.
- B
Set the 'class_weight' parameter in your model to give higher importance to the minority class.
- C
Use a different SageMaker instance type with more GPU capacity to handle the imbalance.
- D
Apply SMOTE (Synthetic Minority Oversampling Technique) to generate synthetic samples for the minority class.
- E
Evaluate the model using metrics like F1-score or precision-recall curve instead of accuracy.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Class imbalance can negatively impact the performance of machine learning models, as standard algorithms tend to favor the majority class. To address this, you can use techniques like oversampling or synthetic data generation (e.g., SMOTE), adjust the model's class weights, and evaluate the model using metrics that focus on minority class performance like F1-score or precision-recall curve. Simply increasing computational resources does not address the data imbalance issue.
- A. Correct.
Correct. Oversampling techniques are commonly used to balance datasets by increasing the number of samples for the minority class, which helps the model learn the minority class better.
- B. Correct.
Correct. Adjusting the 'class_weight' parameter is a standard approach to penalize misclassifications of the minority class more heavily, helping to address class imbalance.
- C. Incorrect.
Incorrect. Increasing GPU capacity does not address class imbalance directly. It may improve training speed but won't solve the underlying issue of imbalanced data.
- D. Correct.
Correct. SMOTE is a widely-used technique to generate synthetic samples for the minority class, effectively balancing the dataset.
- E. Correct.
Correct. Metrics like F1-score or precision-recall curve are better suited for imbalanced datasets since they consider the performance of the minority class rather than accuracy, which can be misleading in such scenarios.