Correct Answer: D. Define problem, acquire and explore data, prepare features, model, evaluate, communicate, and monitor
Explanation:
An end-to-end case study connects problem formulation with data work, modeling, evaluation, communication, and use. The order may iterate, but all major stages should be addressed.
Correct Answer: D. Data sources, transformation steps, code versions, parameters, and evaluation results
Explanation:
Reproduction requires a record of how raw inputs were transformed into final outputs. Versions and parameters are essential because small changes can alter results.
Correct Answer: D. It collects outcome information that can improve future evaluation and retraining
Explanation:
Feedback links predictions or recommendations to subsequent user responses and outcomes. Those observations help determine whether the system is effective and how it should be updated.
Correct Answer: D. Data or concept drift requiring monitoring
Explanation:
Drift occurs when input distributions or relationships between inputs and outcomes change after deployment. Monitoring is needed to detect deterioration and trigger review or retraining.
Correct Answer: A. Making a trained model available within an operational system or decision process
Explanation:
Deployment connects the model to real inputs so its predictions can be used in practice. It may involve an application, API, batch job, or embedded business workflow.
Correct Answer: B. Findings at later stages often require revisiting earlier assumptions, data, or features
Explanation:
EDA, modeling, and evaluation frequently reveal issues that require new cleaning, data collection, or reformulation. The workflow therefore moves back and forth rather than following a rigid one-way path.
Correct Answer: D. To verify that the added complexity produces meaningful improvement
Explanation:
A baseline provides a minimum reference such as predicting the mean or majority class. A sophisticated model is worthwhile only if it improves on that reference sufficiently.
Leakage occurs when predictors contain information unavailable at the actual time of prediction. It creates unrealistically high evaluation results that will not be reproduced in use.
Correct Answer: A. Selecting models or hyperparameters before final test evaluation
Explanation:
Validation data guide choices such as model type, complexity, or hyperparameter values. The test set should remain untouched until those choices are complete.
