MLS-C01 Question 226
Select 3You are developing a machine learning model to predict customer churn for a subscription-based service. After training, you notice that the model achieves high accuracy on the training data but performs poorly on the validation data, and its predictions favor one customer demographic over others. Which of the following strategies could help address both the model's variance and bias issues?
- A
Increase the size of the training dataset by collecting more diverse samples.
- B
Perform feature scaling to ensure that all input features are on a similar scale.
- C
Use regularization techniques, such as L2 regularization, to reduce overfitting.
- D
Perform a bias audit and re-sample the dataset to balance underrepresented groups.
- E
Use a more complex model architecture to improve the model’s capacity to learn patterns.
Show answer and explanation
Correct answers: A, C, D
Explanation
To address high variance, strategies like increasing the dataset size and applying regularization are effective. To mitigate bias, conducting a bias audit and ensuring balanced representation in the dataset are essential steps. Using a more complex model architecture, however, could exacerbate overfitting and worsen variance issues, making it an unsuitable approach for this scenario.
- A. Correct.
Increasing the size of the training dataset, especially with diverse samples, can help reduce variance and improve generalization to the validation set. It also introduces more representative data, reducing bias toward specific groups.
- B. Incorrect.
While feature scaling is useful for optimizing the training process, it does not directly address bias or variance issues. It is more relevant for models sensitive to feature magnitudes, such as logistic regression or neural networks.
- C. Correct.
Regularization techniques like L2 regularization help reduce overfitting, which directly addresses high variance by penalizing overly complex models.
- D. Correct.
A bias audit and re-sampling the dataset to balance underrepresented groups can directly address bias in the training data and improve fairness in predictions.
- E. Incorrect.
Using a more complex model architecture can increase the risk of overfitting and worsen variance, especially if the training dataset is not large or diverse enough.