Google Professional Machine Learning Engineer Question 512
Select 3Google Cloud PlatformYou are tasked with designing a machine learning model on Google Cloud to predict customer churn. The dataset contains over 1,000 features, many of which are highly correlated or irrelevant. What are the best steps to reduce the feature dimensions while maintaining the predictive accuracy of the model?
- A
Apply Principal Component Analysis (PCA) to reduce the number of features while preserving maximum variance in the data.
- B
Use a feature selection technique like Recursive Feature Elimination (RFE) to identify and retain the most relevant features.
- C
Use AutoML Tables in Google Cloud, which automatically handles feature selection and dimensionality reduction.
- D
Manually drop features based on domain knowledge without any further analysis.
- E
Increase the size of your dataset to better handle the high dimensionality.
Show answer and explanation
Correct answers: A, B, C
Explanation
Dealing with high-dimensional datasets requires effective dimensionality reduction and feature selection techniques to improve model performance and reduce complexity. Principal Component Analysis (PCA) reduces the number of features by transforming them while retaining variance. Recursive Feature Elimination (RFE) systematically selects the most relevant features. AutoML Tables simplifies this process by automating feature selection and dimensionality reduction. Manual elimination of features without analysis is unreliable, and increasing dataset size does not address the issue of redundant or irrelevant features.
- A. Correct.
Correct: Principal Component Analysis (PCA) is a common dimensionality reduction technique that transforms features into a smaller set while retaining maximum variance, which can help with high-dimensional data.
- B. Correct.
Correct: Recursive Feature Elimination (RFE) is a feature selection technique that iteratively removes less important features, helping to reduce dimensionality while maintaining model accuracy.
- C. Correct.
Correct: AutoML Tables in Google Cloud automatically performs feature selection and dimensionality reduction as part of its training process, making it a practical choice for this use case.
- D. Incorrect.
Incorrect: While domain knowledge is valuable, relying solely on manual feature elimination without analysis can lead to suboptimal results and missed opportunities to leverage useful features.
- E. Incorrect.
Incorrect: Increasing the dataset size can help mitigate overfitting in high-dimensional data but does not directly address dimensionality reduction or feature selection.