Hands-On Exercise: Demonstrate Overfitting and Underfitting CT-AI Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A team is building a decision-tree classifier for a medical dataset with 200 features and only 1,000 patient records. They set the tree depth to 20 with no minimum samples per leaf. The model achieves 100% accuracy on training data but 65% on the test set. Which combination of changes is most likely to improve test performance?

   • AIncrease the tree depth to 30 and remove the minimum samples per leaf constraint.
   • BReduce the number of features to 5 and keep the depth at 20.
   • CKeep the depth at 20 but remove the minimum samples per leaf to allow finer splits.
   • DReduce the tree depth, increase the minimum samples per leaf, and apply pruning.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   During training, you notice that the training loss continues to decrease while the validation loss starts to increase. What does this indicate?

   • AThe model is overfitting.
   • BThe learning rate is too low.
   • CThe model is underfitting.
   • DThe model is learning correctly and will soon generalize better.
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   In a hands-on exercise to demonstrate overfitting, what would you expect to observe in the training and test errors?

   • ATraining error is high, while test error is low.
   • BBoth training and test errors are high.
   • CTraining error is very low, while test error is high.
   • DBoth training and test errors are low and similar.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   Which scenario is a typical example of underfitting?

   • AA model is trained on a large dataset with regularization, achieving similar performance on training and test sets.
   • BA deep neural network with many layers is trained on a small dataset and achieves near-perfect accuracy on training data but fails on test data.
   • CA linear model is used to fit data that has a clear quadratic relationship, and the model performs poorly on both training and test data.
   • DA model is trained with a very high learning rate, causing its loss to oscillate and never converge.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A team trains a support vector machine with a linear kernel on a dataset where the decision boundary is clearly circular. The model achieves 55% accuracy on both training and test sets. Which statement best describes the situation?

   • AThe model is overfitting because it fails to capture the circular boundary.
   • BThe model is underfitting because a linear kernel cannot represent the circular decision boundary.
   • CThe model is correctly fitted because 55% accuracy is typical for SVMs.
   • DThe result is inconclusive without trying a different kernel.
   Reveal Solution Discussion Select an answer first