AI Model Training, Testing, and Validation AAIR Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   Which testing strategy is most appropriate for evaluating a model's accuracy on a specific subset of data, such as edge cases?

   • ATesting only on the training data
   • BTargeted testing using curated test cases that cover edge scenarios
   • CUsing a single large test set without any specific focus
   • DRandom sampling of the entire dataset
   Reveal Solution Discussion Select an answer first
2. 27expert · hard

   A credit-card company is developing a fraud-detection model. The team observes that a simple logistic regression model has high bias but low variance, while a deep neural network has low bias but high variance. The risk team needs a model that performs consistently on new data. Which approach best balances bias and variance?

   • AAverage the predictions of the logistic regression and neural network models
   • BAlways choose the logistic regression model because it has low variance
   • CAlways choose the neural network because it has low bias
   • DUse a model of intermediate complexity, such as a gradient-boosted tree, and tune it using cross-validation
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   What does precision measure in a classification model?

   • AThe proportion of actual positives that are correctly identified
   • BThe overall correctness of the model across all classes
   • CThe ratio of true negatives to all negative predictions
   • DThe proportion of positive predictions that are actually correct
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   In k-fold cross-validation, what happens to the dataset?

   • AIt is randomly partitioned into k equal-sized folds
   • BIt is duplicated k times to create more training data
   • CIt is split into a training set and a test set only
   • DIt is sorted by target value to ensure balanced folds
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A retail company is building a recommendation model. The data science team has split the data into training, validation, and test sets. They are now tuning hyperparameters. Which set should they use to make hyperparameter tuning decisions?

   • AA combination of training and test sets, to maximize the data used for tuning
   • BThe test set, to get the most accurate estimate of real-world performance
   • CThe validation set, to select the best hyperparameters without biasing the test set
   • DThe training set, to ensure the model fits the data well
   Reveal Solution Discussion Select an answer first