Objective 4.4 Address Business Risks, Ethical Concerns, and Related Concepts in Operationalizing the Model AIP-210 Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A global insurance company is deploying a claims fraud detection model. The model uses customer data from multiple countries. The company must comply with GDPR in Europe and other local data protection laws. The data science team wants to use a centralized model trained on all data. What is the most appropriate approach?

   • AObtain consent from all customers to use their data in the centralized model.
   • BAnonymize all data before training the centralized model.
   • CUse a federated learning approach where the model is trained across regions without moving raw data.
   • DTrain separate models for each region to avoid cross-border data transfers.
   Reveal Solution Discussion Select an answer first
2. 2expert · hard

   A hospital's ML model for predicting patient deterioration is deployed. Six months later, the model's performance degrades because the patient population has changed (e.g., more elderly patients). The hospital's clinical staff are concerned about patient safety. What is the best way to manage this risk?

   • ASet up a monitoring system that tracks model performance and data drift, and triggers a review when performance drops below a threshold.
   • BIncrease the frequency of model retraining to monthly, regardless of performance.
   • CReplace the model with a simpler rule-based system that is easier to understand.
   • DRetrain the model immediately with the most recent patient data.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   A financial services company deploys an ML model to approve personal loans. Which of the following is an example of a financial business risk associated with this deployment?

   • AThe model's training data contains historical biases against certain demographic groups.
   • BThe model's predictions are not easily explainable to regulators.
   • CThe model's decisions are not transparent to customers, reducing trust.
   • DThe model incorrectly approves high-risk loans, leading to increased default losses.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A model deployment poses a reputational risk due to lack of transparency. Which of the following is a mitigation strategy?

   • ADeploy the model without any documentation to avoid scrutiny.
   • BLimit the model's usage to internal teams only.
   • CIncrease the model's training data to improve performance.
   • DImplement explainable AI techniques to provide insights into model decisions.
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A financial services company is developing a model to detect money laundering. The model has a high false-positive rate, causing investigators to waste time on false leads. The company is also under regulatory pressure to improve detection. The data science team proposes a more complex model that may reduce false positives but is less interpretable. What is the best approach?

   • AUse an ensemble of models and average their predictions to reduce false positives.
   • BKeep the current model to maintain interpretability, even though it has a high false-positive rate.
   • CDeploy the complex model and use explainability tools to provide justifications for each alert.
   • DIncrease the decision threshold to reduce false positives, even if it means missing some true positives.
   Reveal Solution Discussion Select an answer first