Underfitting CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A team is building a model to predict customer lifetime value. They have a dataset with 50 features. They train a linear regression model and get an R² of 0.30 on both training and test sets. They then add all pairwise interaction terms, increasing the feature count to 1,275. The training R² jumps to 0.95, but the test R² drops to 0.20. What is the most appropriate next step?

   • AAdd L1 regularization to the model with all interaction terms to force some weights to zero.
   • BRevert to the original linear model because the interaction terms caused overfitting.
   • CUse a moderate number of interaction terms or a model with built-in regularization, such as ridge regression, to balance bias and variance.
   • DKeep all interaction terms because the training R² is high, indicating the model is learning well.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A team is training a k-nearest neighbors (KNN) classifier with k=50. The model achieves 55% accuracy on both training and test sets. The team reduces k to 5, and accuracy improves to 85% on both sets. What was the problem with the original model?

   • AThe model was suffering from data leakage because the test accuracy was too low.
   • BThe model was underfitting because a large k value makes the decision boundary too smooth and simple.
   • CThe model was well-fitted, and the improvement was due to random chance.
   • DThe model was overfitting because a large k value memorizes the training data.
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A team is training a decision tree to predict customer churn. The tree has a maximum depth of 2. The model achieves 65% accuracy on both training and test sets. The team has plenty of data and the features are known to be predictive. What is the most likely cause of the poor performance?

   • AThe model is too simple because the maximum depth is too restrictive, preventing it from learning complex patterns.
   • BThe model is overfitting because the tree is memorizing the training data.
   • CThe model is suffering from data leakage because the test accuracy is lower than the training accuracy.
   • DThe model needs more training data to learn the patterns.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   An ML engineer is training a gradient boosting model. The training error is decreasing steadily, but the validation error has plateaued and is slightly increasing. The engineer suspects overfitting and decides to add more regularization. However, after adding strong regularization, both training and validation errors increase significantly. What is the most likely explanation for this outcome?

   • AThe learning rate was too high, causing the model to skip the optimal solution.
   • BThe model was actually underfitting, and the regularization reduced its already-insufficient capacity.
   • CThe model was overfitting, and the regularization successfully reduced variance, but the validation error increased due to noise.
   • DThe regularization was too strong and caused the model to overfit the validation set.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   An ML engineer plots the learning curve for a classification model. The training error is high and nearly flat, and the validation error is also high and nearly flat, with a small gap between them. As the training set size increases, both curves remain flat. What does this learning curve indicate?

   • AThe model needs more data; the flat curves indicate that adding more data will reduce the error.
   • BThe model is overfitting; the small gap between the curves indicates the model is memorizing the training data.
   • CThe model is underfitting; both errors are high and the curves converge, showing the model cannot learn the pattern.
   • DThe model is well-fitted; the small gap between the curves indicates good generalization.
   Reveal Solution Discussion Select an answer first