Explaining and monitoring a tabular credit risk model on Vertex AI

PMLE By islever Editorial Last updated

Short answer

A Use Vertex Explainable AI sampled Shapley to attribute loan decisions to features, and enable Vertex AI Model Monitoring for feature distribution drift over time.

Question

You work at a bank. You need to develop a credit risk model to support loan application decisions. You decide to implement the model by using a neural network in TensorFlow. Due to regulatory requirements, you need to be able to explain the model’s predictions based on its features. When the model is deployed, you also want to monitor the model’s performance over time. You decided to use Vertex AI for both model development and deployment. What should you do?
  1. Use Vertex Explainable AI with the sampled Shapley method, and enable Vertex AI Model Monitoring to check for feature distribution drift. Correct Answer
  2. Use Vertex Explainable AI with the sampled Shapley method, and enable Vertex AI Model Monitoring to check for feature distribution skew.
  3. Use Vertex Explainable AI with the XRAI method, and enable Vertex AI Model Monitoring to check for feature distribution drift.
  4. Use Vertex Explainable AI with the XRAI method, and enable Vertex AI Model Monitoring to check for feature distribution skew.

Key Takeaways

The question tests matching the explainability method to data type (sampled Shapley for tabular, XRAI for images) and distinguishing training-serving skew from drift detected over time.

For a tabular credit risk model, use Vertex Explainable AI with the sampled Shapley method and enable Vertex AI Model Monitoring to detect feature distribution drift over time (A).

Common mistake: Choosing B or D: feature distribution skew describes a training-versus-serving mismatch at a point in time, not the ongoing change 'over time' that the scenario asks to monitor, which is drift.

Expert Analysis

The two requirements split cleanly. First, the model must be explained from its features. The loan data here is tabular — income, debt-to-income ratio, credit history — so the attribution method that fits is sampled Shapley, which assigns each feature a contribution to a specific prediction. That is what a regulator wants to hear when asking "why was this applicant declined?"

Why XRAI is wrong (options C and D). XRAI — Explainable AI for Images — segments an image into regions and attributes the prediction to those regions. It only applies to image classification inputs. A TensorFlow credit risk network consuming structured features gives XRAI nothing to segment, so C and D are eliminated immediately. The tell in the question is that it describes features such as loan attributes rather than pixels or image regions.

Drift versus skew — the monitoring boundary. This is the concept the question hinges on. Training-serving skew compares the serving-time feature distribution against the training-time baseline and typically shows up right after deployment or after a schema/data-source change; it is a static before/after comparison. Drift compares the current serving distribution against a previously observed serving distribution over successive time windows, capturing gradual change as the population and economy evolve. The question says "monitor the model's performance over time," which is precisely the drift use case — a bank's applicant pool shifts as economic conditions change. So the correct pairing is sampled Shapley plus feature distribution drift, which is option A.

A note on the community reasoning: one comment dismisses skew merely because "drift is generally more significant for credit risk," which is not the real justification. The real discriminator is the phrase over time; skew would be the better answer only if the question asked about a mismatch between training and serving data.

Official Reference

Exam Strategy

On Vertex AI explainability questions, first classify the input data type: images point to XRAI, tabular or structured features point to sampled Shapley. Then read the monitoring wording carefully — 'over time' signals drift, while training-versus-serving comparison signals skew. Eliminating on the data type alone usually narrows the choices to two, so the time wording becomes the deciding phrase.

Frequently Asked Questions

When should I use sampled Shapley instead of XRAI in Vertex Explainable AI?

Use sampled Shapley for tabular or structured features such as income and credit history; it attributes a prediction to each feature. XRAI is only for image classification, where it highlights image regions driving the prediction.

What is the difference between feature distribution drift and skew in Vertex AI Model Monitoring?

Skew compares serving-time feature distributions against the training baseline, typically at deployment. Drift compares current serving distributions against earlier serving distributions across time windows, catching gradual change.

Why does a credit risk model need both explainability and monitoring?

Regulators require per-application feature attributions from sampled Shapley, while monitoring over time detects drift in applicant features that could silently degrade loan approval accuracy.

More PMLE FAQ →

Community Votes

A
100%

100% of anonymous learners picked answer A. Votes are pick records left by other test-takers — they are not the verified answer.

Discussion 5 contributions

Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.

From public community threads 5

  1. b1a8fae 👍 10 Answered A
    Not image -> not XRAI Performance over time -> drift, not skew
  2. fitri001 👍 3 Answered A
    why not the others? B. Feature Distribution Skew: While skew can be relevant, drift is generally a more significant concern for credit risk models. Drift indicates a change in the underlying data distribution, potentially impacting model performance. C & D. XRAI Method: XRAI (Explainable AI for Images) is specifically designed for interpreting image classification models. It wouldn't be the most effective choice for a neural network-based credit risk model working with tabular data.
  3. winston9 👍 2 Answered A
    Explainable AI with the XRAI method is for unstructured, image region analysis, in this case we use structured data for loan approval analysis.
  4. taylz876 👍 1 Answered A
    There are two decisions, one for each requirement. • Explain the predictions based on the features: a credit risk model uses tabular features such as income, debt and credit history. Sampled Shapley is the feature attribution method for this kind of data, and it shows how much each feature pushed a prediction up or down, which is what regulators want. XRAI only works on images. • Monitor performance over time: Google's wording is that drift is a change over time, while skew is a mismatch at any point, including the start. The question asks about performance over time, so it's drift. Drift monitoring compares recent production data against earlier production data. Why the others are wrong • C and D: XRAI produces region-level explanations for images. A credit risk model has no pixels, so XRAI doesn't apply. • B and D: Skew compares production data against the training data. The question asks about change over time after deployment, which is drift.
  5. fitri001 👍 1
    Vertex Explainable AI: This is a built-in Vertex AI feature that helps understand how features contribute to model predictions. Sampled Shapley Method: This is a well-suited method for explaining complex models like neural networks. It provides insights into feature importance without requiring retraining the entire model.

Reader comments & corrections

No reader corrections yet — spotted an error or have a note? Share it below.

Log in to comment, report an error, or add a note about this question.

Practice All PMLE Questions

Access 65 questions with complete answers and detailed explanations.

View Full PMLE Practice Test →

← Back to PMLE Study Guide