Google Professional Machine Learning Engineer Question 22
Select 2Google Cloud PlatformYou are building a machine learning model in BigQuery ML to predict customer churn for a subscription-based service. Your dataset contains customer demographics, subscription details, and engagement metrics. During the feature engineering phase, you need to improve model performance by selecting the most relevant features for your model. Which actions can you take to achieve this in BigQuery ML?
- A
Use the ML.FEATURE_INFO function to identify and remove features with low feature importance.
- B
Perform feature transformation using SQL functions such as LOG, SQRT, or NORMALIZE to preprocess numerical features.
- C
Manually remove features with high correlation after visualizing them using a heatmap generated outside of BigQuery.
- D
Use the ML.PREDICT function to iteratively test different feature subsets and evaluate model performance.
- E
Apply dimensionality reduction techniques, such as Principal Component Analysis (PCA), directly in BigQuery ML.
Show answer and explanation
Correct answers: A, B
Explanation
Feature engineering and selection in BigQuery ML can be done using built-in functions such as ML.FEATURE_INFO for feature importance and SQL-based transformations like LOG and NORMALIZE. These tools help improve model performance by identifying relevant features and preprocessing numerical data. While other techniques such as correlation analysis or PCA are valid in general, they are not directly supported within BigQuery ML.
- A. Correct.
Correct. The ML.FEATURE_INFO function in BigQuery ML provides feature importance scores for models, allowing you to identify and remove features with low importance to improve performance.
- B. Correct.
Correct. BigQuery ML supports SQL-based feature transformations (e.g., LOG, SQRT, NORMALIZE) to preprocess numerical features, which can improve model performance by scaling or normalizing data.
- C. Incorrect.
Incorrect. While visualizing high correlations using heatmaps is a valid technique, it requires external tools, and the question focuses on actions within BigQuery ML. Manual removal in this way is not directly supported in BigQuery ML.
- D. Incorrect.
Incorrect. The ML.PREDICT function is used for generating predictions on new data, not for selecting or evaluating feature subsets. This is not an appropriate method for feature selection in BigQuery ML.
- E. Incorrect.
Incorrect. BigQuery ML does not natively support dimensionality reduction techniques like PCA. These techniques would require external tools or libraries outside of BigQuery ML.