AI Capabilities, Data Dependencies, and Failure Modes CAIPM Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   A university wants to use AI to predict student dropout risk based on academic records, attendance, and demographic data. The dataset is small (only 500 students) and heavily imbalanced (only 5% dropouts). The model trained on this data shows high accuracy but fails to identify actual dropouts. What is the most likely cause and the best mitigation?

   • AData quantity and class imbalance; use oversampling or synthetic data generation to balance the classes and collect more data.
   • BData relevance; the demographic data is irrelevant, so remove it from the model.
   • CModel overfitting; the model is too complex, so simplify it.
   • DData quality; the attendance records contain errors, so clean the data.
   Reveal Solution Discussion Select an answer first
2. 37application · medium

   An e-commerce company's product recommendation model performs very well on historical customer data but fails to recommend relevant products for new customers. The model was trained on a large dataset of past purchases. What is the most likely cause of this failure?

   • AThe model lacks natural language processing capabilities to understand product descriptions.
   • BThe model is suffering from data drift because customer preferences change over time.
   • CThe model is biased against new customers because of their demographic attributes.
   • DThe model is overfitting to the historical data and cannot generalize to new customers.
   Reveal Solution Discussion Select an answer first
4. 38expert · hard

   A social media company uses an AI system to detect hate speech. The system is trained on public posts, but it performs poorly on posts from certain dialects. The company has a limited budget and must improve performance for those dialects. Which approach is most effective?

   • ATranslate all posts to a standard language before classification.
   • BUse a larger pre-trained language model to improve understanding of dialects.
   • CCollect and label more data from those dialects and retrain the model.
   • DIncrease the model's sensitivity threshold to catch more hate speech.
   Reveal Solution Discussion Select an answer first
5. 39expert · hard

   A healthcare AI system predicts patient deterioration and alerts nurses. The system was trained on data from a specific hospital and performs well there. The system is being expanded to a rural hospital with a different patient demographic and limited IT staff. The rural hospital cannot support frequent manual retraining. What is the best strategy to maintain model performance?

   • AUse transfer learning to adapt the model to the rural hospital's data with minimal retraining, and implement automated monitoring with alerts for drift.
   • BRetrain the model from scratch using only the rural hospital's data.
   • CDeploy the model as-is and rely on nurses to report errors.
   • DUse a simpler model that requires no retraining and is robust to demographic differences.
   Reveal Solution Discussion Select an answer first
6. 40application · medium

   A transportation company uses an AI model to predict traffic congestion. The model was trained on historical traffic data from the past five years. Recently, a major highway was rerouted, and the model's predictions have become inaccurate. What is the most effective way to address this?

   • AAdjust the model's threshold to reduce false predictions.
   • BRetrain the model with new traffic data that includes the rerouted highway.
   • CUse a simpler model that is less sensitive to changes.
   • DFreeze the model to maintain consistent predictions.
   Reveal Solution Discussion Select an answer first