AI in Customer Churn Prediction Quiz

Answers and explanations appear as you take the quiz.

1. What does 'customer churn' refer to?
   • Customers canceling or ceasing to use a product or service
   • Customers increasing their spending
   • The process of onboarding new customers
   • A type of marketing email
2. What type of machine learning problem is churn prediction usually framed as?
   • Unsupervised clustering
   • Supervised classification using historical labeled data
   • Reinforcement learning from rewards
   • Image generation
3. Why is plain accuracy a poor metric for churn models?
   • Accuracy is impossible to compute
   • Accuracy only works for images
   • Churn datasets are imbalanced, so predicting 'no churn' for everyone can look highly accurate
   • Churn has no measurable outcome
4. What is 'data leakage' in this context?
   • Customers sharing their passwords
   • Losing customer records to hackers
   • Slow database queries
   • A feature that accidentally encodes the outcome, like a cancellation date, inflating performance
5. Which algorithm family is especially popular for tabular churn data?
   • Gradient-boosted decision trees like XGBoost and LightGBM
   • Convolutional neural networks
   • Recurrent neural networks for audio
   • K-means clustering only
6. What does 'uplift modeling' add beyond standard churn prediction?
   • It makes the model run faster
   • It estimates which customers may be more likely to benefit from an intervention
   • It removes the need for any data
   • It only works for new customers
7. Which feature group can help a churn model detect weakening customer engagement?
   • The color of the customer’s profile image
   • A feature recorded only after cancellation has occurred
   • Recent activity and usage frequency, alongside contract and support history
   • The model’s GPU temperature during training
8. What should a team define when labeling a churn model’s prediction target?
   • The color scheme of the account dashboard
   • The number of features before collecting customer history
   • A fixed decision threshold before model training begins
   • A clear prediction horizon, such as whether a customer will leave within the next 30 or 90 days