Objective 3.5 Address Business Risks, Ethical Concerns, and Related Concepts in Training and Tuning AIP-210 Practice Questions (Page 6)

Questions 26–30

1. 26application · medium

   A financial services firm is developing a model to detect fraudulent transactions. The model will be used in production and must be auditable by regulators. The firm wants to ensure that every model version is documented, including its training data, hyperparameters, and performance metrics. Which practice should they adopt?

   • ARely on the data scientist's memory to explain the model during audits.
   • BKeep the final model file in a shared folder and overwrite it whenever a new version is trained.
   • CMaintain a model registry that records metadata for each version, including data lineage and evaluation results.
   • DDocument the model architecture in a wiki page and update it informally.
   Reveal Solution Discussion Select an answer first
2. 27application · medium

   A social media company wants to train a sentiment analysis model on user posts. The data includes user comments that may contain personal information. The company operates in multiple jurisdictions with different privacy laws. They need to ensure that the model training complies with the strictest applicable regulation. What should they do?

   • AUse only publicly available posts to avoid privacy issues.
   • BAnonymize all user data before training and document the anonymization process.
   • CStore the data on servers located in the country with the most lenient privacy laws.
   • DTrain the model only on data from users who have consented to data use.
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   Which of the following is a model-agnostic explainability technique that approximates a model's predictions locally?

   • ACross-validation
   • BLIME (Local Interpretable Model-agnostic Explanations)
   • CDecision tree pruning
   • DGradient descent
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   An insurance company has deployed a model to set premiums. After a year, they notice that the model's predictions are becoming less accurate because customer behavior has changed. They also observe that the model is now approving riskier customers than before. What should the company do to address this?

   • AIncrease the model's complexity to adapt to the new customer behavior.
   • BReduce the model's learning rate to stabilize predictions.
   • CSet up automated monitoring for data drift and model performance, and retrain the model when drift exceeds a threshold.
   • DDisable the model and rely on manual underwriting.
   Reveal Solution Discussion Select an answer first
6. 30foundation · easy

   A company is training a model on data from California residents. Which regulation is most likely to apply?

   • APIPEDA (Personal Information Protection and Electronic Documents Act)
   • BGDPR
   • CCCPA (California Consumer Privacy Act)
   • DLGPD (Lei Geral de Proteção de Dados)
   Reveal Solution Discussion Select an answer first