Describe Core Machine Learning Concepts AI-900 Practice Questions (Page 5)

Questions 21–25

1. 21expert · hard

   A healthcare research team is developing a model to predict patient outcomes. The dataset contains records from multiple hospitals collected over five years. The team must ensure the model generalizes to new hospitals while also detecting data drift over time. They have limited compute resources and can only train the model once. Which splitting strategy best balances these constraints?

   • AUse a group-based split that holds out entire hospitals and uses the most recent data from those hospitals for validation
   • BSplit chronologically so validation contains the most recent data
   • CRandomly split the data into training and validation sets
   • DSplit by hospital so that validation contains hospitals not seen in training
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   What is the main purpose of a validation dataset in machine learning?

   • ATo provide an unbiased final evaluation of the model's performance
   • BTo evaluate the model during training and help tune hyperparameters
   • CTo store the model's predictions for comparison with labels
   • DTo train the model's parameters
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A team is training a spam email classifier. They have 100,000 labeled emails. They split the data into training and validation sets. After training, they use the validation set to measure accuracy. Why is it important to use a separate validation set rather than evaluating on the training set?

   • ATo reduce the time required to train the model
   • BTo ensure the model memorizes the training data
   • CTo get an unbiased estimate of how the model will perform on new emails
   • DTo increase the size of the training data
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A data scientist is training a model to predict equipment failure. The dataset contains sensor readings recorded every minute for 1,000 machines over a year. Each machine has multiple failure events. The scientist wants to evaluate the model's ability to predict failures on machines not seen during training. Which splitting approach is most appropriate?

   • ASplit by time so that validation contains the last month of readings
   • BSplit by machine so that validation contains machines not in training
   • CUse all data for training and validate on the same data
   • DRandomly split all rows into training and validation sets
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A retail company wants to train a model to predict whether a customer will make a purchase within the next 30 days. The dataset includes customer age, past purchase amount, number of website visits, and a column indicating whether a purchase was made. The company plans to use the model to score new customers. Which column should be used as the label?

   • ANumber of website visits
   • BWhether a purchase was made
   • CCustomer age
   • DPast purchase amount
   Reveal Solution Discussion Select an answer first