AI Model Training, Testing, and Validation AAIR Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A data science team is developing a model with a small dataset of 1,000 samples. They are concerned about the model's stability across different data subsets. Which approach provides the most robust assessment of model stability?

   • AUse k-fold cross-validation and report the mean and standard deviation of the performance metric across folds
   • BUse a single 80/20 train-test split and report the test accuracy
   • CTrain the model on the entire dataset and report the training accuracy
   • DUse a 50/50 train-test split and compare the training and test accuracy
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   An insurance company is deploying a risk-scoring model. The team observes that increasing model complexity improves validation-set performance but degrades test-set performance. The compliance officer requires a model that generalizes reliably to new applicants. Which decision best balances the bias-variance trade-off?

   • AUse the simplest possible model to guarantee low variance and stable predictions
   • BAverage predictions from all model complexities to cancel out individual errors
   • CChoose the most complex model because it fits the validation data best
   • DSelect the model complexity that minimizes validation-set error, then confirm performance on the test set
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   What is the primary goal of bias and fairness testing in AI models?

   • ATo reduce the model's complexity
   • BTo increase the speed of model inference
   • CTo identify and measure disparate impacts on different demographic groups
   • DTo ensure the model has high accuracy on the training set
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   In the context of AI model training, what is the primary purpose of the iterative optimization phase?

   • ATo permanently freeze the model's architecture to prevent further changes
   • BTo select the algorithm that best fits the business problem
   • CTo increase the size of the training dataset to improve model accuracy
   • DTo adjust model parameters to minimize the loss function on the training data
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A logistics company is training a model to predict delivery delays. The raw data includes timestamps, weather, traffic, and driver IDs. The team notices that the model performs very well on training data but poorly on new routes. Which data-preparation practice would most likely improve generalization?

   • ARemove all weather and traffic features because they add noise
   • BUse only the most recent month of data for training to reflect current conditions
   • CIncrease the model's complexity to capture more nuances in the training data
   • DEngineer features that capture route-level patterns rather than relying on driver-ID-specific behavior
   Reveal Solution Discussion Select an answer first