Objective 3.3 Train, Validate, and Test Data Subsets AIP-210 Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   What is data leakage in the context of data splitting?

   • AUsing too much data for training
   • BLosing data due to a system failure
   • CInformation from the test set influencing the training process
   • DSharing data with external parties
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   How should split ratios be adjusted for very large datasets?

   • AUse equal thirds for all subsets
   • BIncrease the test set to 50% to ensure robust evaluation
   • CAlways use 50% for training to ensure the model sees enough data
   • DUse a smaller percentage for validation and test because absolute sample sizes can still be sufficient
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   What is the primary purpose of the test subset in a machine learning project?

   • ATo augment the training data with additional samples
   • BTo assess the final model's generalization performance on unseen data
   • CTo fit the model's parameters
   • DTo tune hyperparameters
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   A data scientist is working on a binary classification problem where 90% of samples belong to class A and 10% to class B. The dataset has 50,000 samples. The scientist wants to split the data into training (80%) and test (20%) sets. Which approach ensures the test set is representative of the original class distribution?

   • ARandomly shuffle the data and take the first 80% for training and the last 20% for test
   • BUse stratified splitting to maintain the 90/10 class ratio in both training and test sets
   • CSort the data by class label and split sequentially
   • DUse a 50/50 split to ensure both classes are equally represented
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   What is the main difference between random and non-random splitting methods?

   • ANon-random splitting is only used for small datasets
   • BRandom splitting always produces better models
   • CRandom splitting assigns samples arbitrarily, while non-random splitting follows a rule
   • DRandom splitting requires stratification, while non-random does not
   Reveal Solution Discussion Select an answer first