AI Life Cycle (Model Selection, Training, and Validation) AAISM Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A model is designed to detect rare diseases from medical images. The dataset has 1% positive cases. The model achieves 99% accuracy, but the recall is only 10%. The clinical team is concerned about missing cases. Which metric should be prioritized to evaluate the model's effectiveness?

   • AAccuracy, because it is the most intuitive and widely understood metric.
   • BF1-score, because it balances precision and recall equally.
   • CPrecision, because it measures how many predicted positives are actually positive.
   • DRecall, because missing a rare disease has severe consequences.
   Reveal Solution Discussion Select an answer first
2. 7application · medium

   A team is training a support vector machine (SVM) with an RBF kernel. The model's training accuracy is 95%, but test accuracy is only 70%. The team wants to improve generalization. Which hyperparameter adjustment is most likely to help?

   • AIncrease the regularization parameter C to reduce training error.
   • BRemove the RBF kernel and use a linear kernel to reduce variance.
   • CDecrease the regularization parameter C to apply stronger regularization.
   • DIncrease the gamma parameter to make the decision boundary more flexible.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   What is the goal of the bias-variance trade-off in model selection?

   • ATo find the model complexity that minimizes total error
   • BTo minimize both bias and variance to zero
   • CTo maximize bias to simplify the model
   • DTo maximize variance to ensure the model fits the training data perfectly
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   In the bias-variance trade-off, what does high variance typically indicate?

   • AThe model is overly sensitive to small fluctuations in the training data
   • BThe model is too simple and fails to capture patterns
   • CThe model has a high error on the training data
   • DThe model's performance is consistent across different training sets
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A healthcare AI team is developing a model to predict patient readmission. They have data from multiple hospitals collected over several years. The team plans to split the data randomly into training and test sets. However, the model's performance drops significantly when deployed at a new hospital. What is the most likely cause?

   • AThe model overfit to the specific hospitals in the training data and does not generalize to new settings.
   • BThe random split caused data leakage because patients may appear in both sets.
   • CThe model was not trained long enough to converge.
   • DThe test set was too small to provide a reliable performance estimate.
   Reveal Solution Discussion Select an answer first