Challenges in Data Preparation CT-AI Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   What is a potential source of bias in data collection that can lead to unfair ML outcomes?

   • AUnderrepresentation of certain groups in the dataset
   • BApplying one-hot encoding to categorical variables
   • CSplitting data into training and test sets
   • DUsing a large sample size
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which technique addresses class imbalance by creating synthetic samples of the minority class?

   • AMin-max scaling
   • BOne-hot encoding
   • CImputation
   • DSMOTE
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   What is the main goal of feature engineering in the data preparation phase?

   • ATo split the data into training and test sets
   • BTo remove all missing values from the dataset
   • CTo increase the number of features without regard to relevance
   • DTo create or select features that improve model accuracy and interpretability
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A data scientist is working on a binary classification problem where the positive class is rare (2% of the data). The scientist decides to use cost-sensitive learning by assigning a higher misclassification cost to the positive class. What is the primary benefit of this approach compared to oversampling?

   • ACost-sensitive learning is always more accurate than oversampling.
   • BCost-sensitive learning automatically balances the dataset, so no other preprocessing is needed.
   • CCost-sensitive learning is easier to implement than oversampling.
   • DCost-sensitive learning does not require modifying the training data, so it avoids the risk of overfitting to synthetic samples.
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A team is building a model to detect rare diseases. The dataset has only 100 positive cases and 10,000 negative cases. They want to split the data into training and test sets. What is the most appropriate splitting strategy to ensure a reliable evaluation of the model's performance on the positive class?

   • AUse a time-based split because the data is temporal.
   • BPerform a random 80/20 split without any stratification.
   • COversample the positive class before splitting to ensure enough positive cases in the test set.
   • DUse stratified sampling to ensure the test set contains the same proportion of positive cases as the original dataset.
   Reveal Solution Discussion Select an answer first