Overfitting CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A team is building a model to classify images of defective parts on a production line. They have only 200 labeled images per class. The team decides to use a very deep convolutional neural network with millions of parameters. What is the most likely outcome?

   • AThe model will perform well because more parameters always improve accuracy.
   • BThe model will generalize well because deep networks are inherently robust.
   • CThe model will overfit because the model complexity is high relative to the amount of training data.
   • DThe model will underfit because the network is too complex for the task.
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A team is building a model to predict housing prices. They have a dataset with 10,000 samples and 500 features. The model is a linear regression with no regularization. The training error is low, but the test error is high. Which of the following is the most effective first step to reduce the test error?

   • AIncrease the model's capacity by using a polynomial regression.
   • BApply feature selection or dimensionality reduction to reduce the number of features.
   • CAdd more features to the model to capture more information.
   • DRemove the test set and use cross-validation only.
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A team is building a model to predict customer churn. They have a dataset with 1,000 samples and 100 features. The model is a logistic regression. They suspect overfitting and are considering using L1 regularization. What is the primary advantage of L1 regularization over L2 regularization in this scenario?

   • AL1 regularization can drive some feature weights to zero, effectively performing feature selection.
   • BL1 regularization is faster to compute than L2 regularization.
   • CL1 regularization reduces overfitting by increasing the model's bias more than L2.
   • DL1 regularization always produces a more accurate model than L2 regularization.
   Reveal Solution Discussion Select an answer first
5. 14foundation · easy

   In the bias-variance tradeoff, what does 'variance' refer to?

   • AThe model's sensitivity to variations in the training data.
   • BThe amount of data used for training the model.
   • CThe error introduced by approximating a complex problem with a simple model.
   • DThe difference between the model's predictions and the true values.
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A data scientist is using k-fold cross-validation to tune a model. They notice that the cross-validation scores vary significantly across folds. What does this indicate?

   • AThe model is performing optimally because cross-validation is always reliable.
   • BThe model is overfitting because it is too sensitive to the specific data in each fold.
   • CThe model has high bias because it cannot generalize across folds.
   • DThe model is underfitting because it is too simple.
   Reveal Solution Discussion Select an answer first