Foundational Concepts DY0-001 Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   In k-fold cross-validation, how is the dataset typically divided?

   • AInto two parts: 70% training and 30% validation
   • BInto k parts, training on all parts simultaneously
   • CInto k equal-sized folds, using one fold for validation and the rest for training in each iteration
   • DInto k parts, using all parts for training and a separate test set for validation
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   In the bias-variance tradeoff, what does high variance typically indicate about a model?

   • AThe model has consistent performance across different datasets.
   • BThe model is unable to learn from the training data.
   • CThe model performs well on training data but poorly on unseen data.
   • DThe model is too simple and fails to capture underlying patterns.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A data scientist is evaluating a k-nearest neighbors (KNN) model. They use a single train/test split (80/20) to evaluate the model and find that the accuracy is 95%. However, when they deploy the model, the accuracy drops to 70%. Which practice would have likely provided a more reliable estimate of model performance?

   • AUsing a different random seed for the train/test split
   • BUsing k-fold cross-validation instead of a single train/test split
   • CUsing accuracy as the evaluation metric
   • DIncreasing the size of the test set to 50%
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A data scientist is training a neural network to predict house prices. The dataset has a few extreme outliers (e.g., a mansion sold for $50M in a neighborhood of $300K homes). The initial model using mean squared error (MSE) as the loss function is heavily influenced by these outliers, resulting in poor predictions for typical homes. Which change to the loss function would make the model more robust to outliers?

   • AIncrease the learning rate of the optimizer
   • BAdd more hidden layers to the network
   • CUse a larger batch size during training
   • DSwitch to mean absolute error (MAE)
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A team is developing a model to detect fraudulent credit card transactions. The dataset is highly imbalanced, with fraud occurring in only 0.1% of cases. A single decision tree achieves high accuracy but poor recall on the minority class. The team needs to improve the recall of the fraud class without drastically reducing precision. Which ensemble approach is most appropriate?

   • AA random forest with a high maximum depth
   • BA voting ensemble that averages the probabilities of all models
   • CA boosting ensemble with a focus on misclassified samples
   • DA standard bagging ensemble of decision trees
   Reveal Solution Discussion Select an answer first