Google Professional Machine Learning Engineer Question 495
Select 3Google Cloud PlatformYour team has built a machine learning model for predicting loan approvals, and stakeholders are concerned about the model's fairness and transparency. You are tasked with improving the explainability of the model using Google Cloud tools. Which of the following approaches would best address this requirement?
- A
Use AI Explanations in Vertex AI to compute feature attributions for model predictions.
- B
Train a simpler linear regression model as a replacement for the current model to improve interpretability.
- C
Enable What-If Tool in Vertex AI to analyze model behavior for specific data points and scenarios.
- D
Use AutoML's built-in explainability features to visualize the influence of features on predictions.
- E
Manually inspect the training data to ensure there are no biases or errors in the dataset.
Show answer and explanation
Correct answers: A, C, D
Explanation
Improving explainability in machine learning models often involves using tools that provide insights into how the model makes decisions. Google Cloud offers several such tools, including AI Explanations in Vertex AI, the What-If Tool, and AutoML's built-in explainability features. These tools allow teams to understand feature attributions, analyze model behavior, and visualize feature importance, directly addressing concerns about transparency and fairness.
- A. Correct.
AI Explanations in Vertex AI provide feature attributions, which help identify how much each feature contributes to a prediction. This is a direct way to improve explainability.
- B. Incorrect.
While simpler models like linear regression are more interpretable, replacing the existing model sacrifices predictive accuracy. This is not the optimal solution if the current model is already performing well.
- C. Correct.
The What-If Tool allows for interactive exploration of model predictions and their sensitivity to feature changes, which is valuable for improving transparency and fairness.
- D. Correct.
AutoML's built-in explainability features provide visualizations of feature importance and can directly enhance model interpretability.
- E. Incorrect.
Inspecting the training data for biases or errors is a good practice for fairness but does not directly address the explainability of the model's predictions.